1berry/qtbase
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

SSE2 optimize QColorSpace CLUT

Browse Source 
Optimize clamping setup for CLUT lookup.

Pick-to: 6.9
Change-Id: I35d65fc00efcb96a54518466e1e1de169571b862
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
(cherry picked from commit 51d5b9e2580a4f54594f616cf9859af86626c887)
Reviewed-by: Qt Cherry-pick Bot <cherrypick_bot@qt-project.org>
This commit is contained in:
Allan Sandfeld Jensen 2024-04-09 15:31:49 +02:00 committed by Qt Cherry-pick Bot
parent 2770bb4927
commit f417776a36
1 changed files with 42 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

42
src/gui/painting/qcolorclut_p.h
View File

@ -16,7 +16,11 @@
//
#include <QtCore/qlist.h>
#include <QtCore/qsimd.h>
#include <QtGui/private/qcolormatrix_p.h>
#if defined(__SSE2__)
#include <immintrin.h>
#endif
QT_BEGIN_NAMESPACE
@ -44,6 +48,43 @@ public:
{
Q_ASSERT(table.size() == qsizetype(gridPointsX * gridPointsY * gridPointsZ * gridPointsW));
QColorVector frac;
#if defined(__SSE2__)
const __m128 minV = _mm_setzero_ps();
const __m128 maxV = _mm_set1_ps(1.0f);
const __m128i gridPointsInt = _mm_loadu_si128(reinterpret_cast<const __m128i *>(&gridPointsX));
const __m128 gridPointsV = _mm_cvtepi32_ps(_mm_add_epi32(gridPointsInt, _mm_set1_epi32(-1)));
__m128 c = _mm_loadu_ps(&v.x);
c = _mm_max_ps(c, minV);
c = _mm_min_ps(c, maxV);
c = _mm_mul_ps(c, gridPointsV);
#if !defined(__SSE4_1__)
const __m128 clo = _mm_cvtepi32_ps(_mm_cvttps_epi32(c)); // truncation == floor for x >= 0
#else
const __m128 clo = _mm_floor_ps(c);
#endif
__m128 chi = _mm_add_ps(clo, maxV);
chi = _mm_min_ps(chi, gridPointsV);
_mm_storeu_ps(reinterpret_cast<float *>(&frac), _mm_sub_ps(c, clo));
const __m128i ilo = _mm_cvtps_epi32(clo);
const __m128i ihi = _mm_cvtps_epi32(chi);
const uint32_t lox = _mm_cvtsi128_si32(ilo);
const uint32_t hix = _mm_cvtsi128_si32(ihi);
#if !defined(__SSE4_1__)
const uint32_t loy = _mm_cvtsi128_si32(_mm_shuffle_epi32(ilo, _MM_SHUFFLE(1, 1, 1, 1)));
const uint32_t loz = _mm_cvtsi128_si32(_mm_unpackhi_epi32(ilo, ilo));
const uint32_t low = _mm_cvtsi128_si32(_mm_shuffle_epi32(ilo, _MM_SHUFFLE(3, 3, 3, 3)));
const uint32_t hiy = _mm_cvtsi128_si32(_mm_shuffle_epi32(ihi, _MM_SHUFFLE(1, 1, 1, 1)));
const uint32_t hiz = _mm_cvtsi128_si32(_mm_unpackhi_epi32(ihi, ihi));
const uint32_t hiw = _mm_cvtsi128_si32(_mm_shuffle_epi32(ihi, _MM_SHUFFLE(3, 3, 3, 3)));
#else
const uint32_t loy = _mm_extract_epi32(ilo, 1);
const uint32_t loz = _mm_extract_epi32(ilo, 2);
const uint32_t low = _mm_extract_epi32(ilo, 3);
const uint32_t hiy = _mm_extract_epi32(ihi, 1);
const uint32_t hiz = _mm_extract_epi32(ihi, 2);
const uint32_t hiw = _mm_extract_epi32(ihi, 3);
#endif
#else
const float x = std::clamp(v.x, 0.0f, 1.0f) * (gridPointsX - 1);
const float y = std::clamp(v.y, 0.0f, 1.0f) * (gridPointsY - 1);
const float z = std::clamp(v.z, 0.0f, 1.0f) * (gridPointsZ - 1);
@ -60,6 +101,7 @@ public:
frac.y = y - static_cast<float>(loy);
frac.z = z - static_cast<float>(loz);
frac.w = w - static_cast<float>(low);
#endif
if (gridPointsW > 1) {
auto index = [&](qsizetype x, qsizetype y, qsizetype z, qsizetype w) -> qsizetype {
return x * gridPointsW * gridPointsZ * gridPointsY