liz/rawTherapee
liz/rawTherapee
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Amaze Demosaic: Speedup, cleaned code, changed nyquist code

Browse Source
This commit is contained in:
heckflosse 2016-01-26 13:10:38 +01:00
parent 2017a0e592
commit ee665d6790
4 changed files with 418 additions and 420 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

769
rtengine/amaze_demosaic_RT.cc
View File

File diff suppressed because it is too large Load Diff

9
rtengine/helpersse2.h
View File

@ -39,8 +39,15 @@ typedef __m128i vint2;
#define STVFU(x,y) _mm_storeu_ps(&x,y) #define STVFU(x,y) _mm_storeu_ps(&x,y)
#endif #endif
static INLINE vfloat LC2VFU(float &a)
{
// Load 8 floats from a and combine a[0],a[2],a[4] and a[6] into a vector of 4 floats // Load 8 floats from a and combine a[0],a[2],a[4] and a[6] into a vector of 4 floats
#define LC2VFU(a) _mm_shuffle_ps( LVFU(a), _mm_loadu_ps( (&a) + 4 ), _MM_SHUFFLE( 2,0,2,0 ) ) vfloat a1 = _mm_loadu_ps( &a );
vfloat a2 = _mm_loadu_ps( (&a) + 4 );
return _mm_shuffle_ps(a1,a2,_MM_SHUFFLE( 2,0,2,0 ));
}
// Store a vector of 4 floats in a[0],a[2],a[4] and a[6] // Store a vector of 4 floats in a[0],a[2],a[4] and a[6]
#if defined(__x86_64__) && defined(__SSE4_1__) #if defined(__x86_64__) && defined(__SSE4_1__)

4
rtengine/rt_math.h
View File

@ -83,8 +83,8 @@ template<typename _Tp>
inline const _Tp intp(const _Tp a, const _Tp b, const _Tp c) { inline const _Tp intp(const _Tp a, const _Tp b, const _Tp c) {
// calculate a * b + (1 - a) * c // calculate a * b + (1 - a) * c
// following is valid: // following is valid:
// intp(a, b+x, c+x) = vintpf(a, b, c) + x // intp(a, b+x, c+x) = intp(a, b, c) + x
// intp(a, b*x, c*x) = vintpf(a, b, c) * x // intp(a, b*x, c*x) = intp(a, b, c) * x
return a * (b-c) + c; return a * (b-c) + c;
} }

22
rtengine/sleefsseavx.c
View File

@ -910,11 +910,20 @@ static INLINE vfloat vnegf(vfloat f) { return (vfloat)vxorm((vmask)f, (vmask)vca
static INLINE vfloat vself(vmask mask, vfloat x, vfloat y) { static INLINE vfloat vself(vmask mask, vfloat x, vfloat y) {
return _mm_blendv_ps(y,x,(vfloat)mask); return _mm_blendv_ps(y,x,(vfloat)mask);
} }
static INLINE vint vselc(vmask mask, vint x, vint y) {
return _mm_blendv_epi8(y,x,mask);
}
#else #else
// three instructions when using SSE2 // three instructions when using SSE2
static INLINE vfloat vself(vmask mask, vfloat x, vfloat y) { static INLINE vfloat vself(vmask mask, vfloat x, vfloat y) {
return (vfloat)vorm(vandm(mask, (vmask)x), vandnotm(mask, (vmask)y)); return (vfloat)vorm(vandm(mask, (vmask)x), vandnotm(mask, (vmask)y));
} }
static INLINE vint vselc(vmask mask, vint x, vint y) {
return vorm(vandm(mask, (vmask)x), vandnotm(mask, (vmask)y));
}
#endif #endif
static INLINE vfloat vselfzero(vmask mask, vfloat x) { static INLINE vfloat vselfzero(vmask mask, vfloat x) {
@ -928,6 +937,16 @@ static INLINE vfloat vselfnotzero(vmask mask, vfloat x) {
return _mm_andnot_ps((vfloat)mask, x); return _mm_andnot_ps((vfloat)mask, x);
} }
static INLINE vint vselizero(vmask mask, vint x) {
// returns value of x if corresponding mask bits are 1, else returns 0
// faster than vselc(mask, x, ZEROV)
return _mm_and_si128(mask, x);
}
static INLINE vint vselinotzero(vmask mask, vint x) {
// returns value of x if corresponding mask bits are 0, else returns 0
// faster than vselc(mask, ZEROV, x)
return _mm_andnot_si128(mask, x);
}
static INLINE vint2 vseli2_lt(vfloat f0, vfloat f1, vint2 x, vint2 y) { static INLINE vint2 vseli2_lt(vfloat f0, vfloat f1, vint2 x, vint2 y) {
vint2 m2 = vcast_vi2_vm(vmaskf_lt(f0, f1)); vint2 m2 = vcast_vi2_vm(vmaskf_lt(f0, f1));
@ -1365,6 +1384,9 @@ static INLINE vfloat vaddc2vfu(float &a)
return _mm_shuffle_ps(a1,a2,_MM_SHUFFLE( 2,0,2,0 )) + _mm_shuffle_ps(a1,a2,_MM_SHUFFLE( 3,1,3,1 )); return _mm_shuffle_ps(a1,a2,_MM_SHUFFLE( 2,0,2,0 )) + _mm_shuffle_ps(a1,a2,_MM_SHUFFLE( 3,1,3,1 ));
} }
static INLINE vfloat vadivapb (vfloat a, vfloat b) {
return a / (a+b);
}
#endif // __SSE2__ #endif // __SSE2__
#endif // SLEEFSSEAVX #endif // SLEEFSSEAVX