whytofly
/
qt6-bb10
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Improve ARGB32ToRGBA64 conversions 

Improves the precision so 255 values map to 65535 exactly.

Change-Id: I366f408e8c6047d52acbed35e9d665249bbaba2b
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
bb10
Allan Sandfeld Jensen 2019-01-31 16:13:16 +01:00
parent 928cab5ff1
commit 60addee938
2 changed files with 29 additions and 35 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
src/gui/painting/qdrawhelper_avx2.cpp
View File

@ -1135,14 +1135,12 @@ static void convertARGBToRGBA64PM_avx2(QRgba64 *buffer, const uint *src, qsizety
const __m256i src1 = _mm256_unpacklo_epi8(srcVector, srcVector);
const __m256i src2 = _mm256_unpackhi_epi8(srcVector, srcVector);
if (!cf) {
dst1 = _mm256_unpacklo_epi8(srcVector, zero);
dst2 = _mm256_unpackhi_epi8(srcVector, zero);
const __m256i alpha1 = _mm256_shuffle_epi8(dst1, shuffleMask);
const __m256i alpha2 = _mm256_shuffle_epi8(dst2, shuffleMask);
dst1 = _mm256_mullo_epi16(dst1, alpha1);
dst2 = _mm256_mullo_epi16(dst2, alpha2);
dst1 = _mm256_add_epi16(dst1, _mm256_srli_epi16(dst1, 7));
dst2 = _mm256_add_epi16(dst2, _mm256_srli_epi16(dst2, 7));
const __m256i alpha1 = _mm256_shuffle_epi8(src1, shuffleMask);
const __m256i alpha2 = _mm256_shuffle_epi8(src2, shuffleMask);
dst1 = _mm256_mulhi_epu16(src1, alpha1);
dst2 = _mm256_mulhi_epu16(src2, alpha2);
dst1 = _mm256_add_epi16(dst1, _mm256_srli_epi16(dst1, 15));
dst2 = _mm256_add_epi16(dst2, _mm256_srli_epi16(dst2, 15));
dst1 = _mm256_blend_epi16(dst1, src1, 0x88);
dst2 = _mm256_blend_epi16(dst2, src2, 0x88);
} else {
@ -1171,14 +1169,12 @@ static void convertARGBToRGBA64PM_avx2(QRgba64 *buffer, const uint *src, qsizety
const __m256i src1 = _mm256_unpacklo_epi8(srcVector, srcVector);
const __m256i src2 = _mm256_unpackhi_epi8(srcVector, srcVector);
if (!cf) {
dst1 = _mm256_unpacklo_epi8(srcVector, zero);
dst2 = _mm256_unpackhi_epi8(srcVector, zero);
const __m256i alpha1 = _mm256_shuffle_epi8(dst1, shuffleMask);
const __m256i alpha2 = _mm256_shuffle_epi8(dst2, shuffleMask);
dst1 = _mm256_mullo_epi16(dst1, alpha1);
dst2 = _mm256_mullo_epi16(dst2, alpha2);
dst1 = _mm256_add_epi16(dst1, _mm256_srli_epi16(dst1, 7));
dst2 = _mm256_add_epi16(dst2, _mm256_srli_epi16(dst2, 7));
const __m256i alpha1 = _mm256_shuffle_epi8(src1, shuffleMask);
const __m256i alpha2 = _mm256_shuffle_epi8(src2, shuffleMask);
dst1 = _mm256_mulhi_epu16(src1, alpha1);
dst2 = _mm256_mulhi_epu16(src2, alpha2);
dst1 = _mm256_add_epi16(dst1, _mm256_srli_epi16(dst1, 15));
dst2 = _mm256_add_epi16(dst2, _mm256_srli_epi16(dst2, 15));
dst1 = _mm256_blend_epi16(dst1, src1, 0x88);
dst2 = _mm256_blend_epi16(dst2, src2, 0x88);
} else {

36
src/gui/painting/qdrawhelper_sse4.cpp
View File

@ -107,28 +107,26 @@ static void convertARGBToRGBA64PM_sse4(QRgba64 *buffer, const uint *src, int cou
for (; i < count - 3; i += 4) {
__m128i srcVector = _mm_loadu_si128((const __m128i *)&src[i]);
if (!_mm_testz_si128(srcVector, alphaMask)) {
if (!_mm_testc_si128(srcVector, alphaMask)) {
if (!RGBA)
srcVector = _mm_shuffle_epi8(srcVector, rgbaMask);
__m128i src1 = _mm_unpacklo_epi8(srcVector, zero);
__m128i src2 = _mm_unpackhi_epi8(srcVector, zero);
bool cf = _mm_testc_si128(srcVector, alphaMask);
if (!RGBA)
srcVector = _mm_shuffle_epi8(srcVector, rgbaMask);
const __m128i src1 = _mm_unpacklo_epi8(srcVector, srcVector);
const __m128i src2 = _mm_unpackhi_epi8(srcVector, srcVector);
if (!cf) {
__m128i alpha1 = _mm_shuffle_epi8(src1, shuffleMask);
__m128i alpha2 = _mm_shuffle_epi8(src2, shuffleMask);
src1 = _mm_mullo_epi16(src1, alpha1);
src2 = _mm_mullo_epi16(src2, alpha2);
alpha1 = _mm_unpacklo_epi8(srcVector, srcVector);
alpha2 = _mm_unpackhi_epi8(srcVector, srcVector);
src1 = _mm_add_epi16(src1, _mm_srli_epi16(src1, 7));
src2 = _mm_add_epi16(src2, _mm_srli_epi16(src2, 7));
src1 = _mm_blend_epi16(src1, alpha1, 0x88);
src2 = _mm_blend_epi16(src2, alpha2, 0x88);
_mm_storeu_si128((__m128i *)&buffer[i], src1);
_mm_storeu_si128((__m128i *)&buffer[i + 2], src2);
__m128i dst1 = _mm_mulhi_epu16(src1, alpha1);
__m128i dst2 = _mm_mulhi_epu16(src2, alpha2);
// Map 0->0xfffe to 0->0xffff
dst1 = _mm_add_epi16(dst1, _mm_srli_epi16(dst1, 15));
dst2 = _mm_add_epi16(dst2, _mm_srli_epi16(dst2, 15));
// correct alpha value:
dst1 = _mm_blend_epi16(dst1, src1, 0x88);
dst2 = _mm_blend_epi16(dst2, src2, 0x88);
_mm_storeu_si128((__m128i *)&buffer[i], dst1);
_mm_storeu_si128((__m128i *)&buffer[i + 2], dst2);
} else {
if (!RGBA)
srcVector = _mm_shuffle_epi8(srcVector, rgbaMask);
const __m128i src1 = _mm_unpacklo_epi8(srcVector, srcVector);
const __m128i src2 = _mm_unpackhi_epi8(srcVector, srcVector);
_mm_storeu_si128((__m128i *)&buffer[i], src1);
_mm_storeu_si128((__m128i *)&buffer[i + 2], src2);
}