RomkaZVO/FC1
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

123

Browse Source
This commit is contained in:
romkazvo
2023-08-07 19:29:24 +08:00
commit 34d6c5d489
4832 changed files with 1389451 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

526
RenderDll/Common/RendElements/FFT_SSE.cpp Normal file
View File

@@ -0,0 +1,526 @@
#include "RenderPCH.h"
// Not for AMD64
#if !defined(WIN64) && !defined(LINUX)
#pragma warning(push)
#pragma warning(disable:4731) // frame pointer register 'ebp' modified by inline assembly code
struct SConstSSE
{
float m_fVal0;
float m_fVal1;
float m_fVal2;
float m_fVal3;
};
_declspec(align(16)) SConstSSE _tabCoef64_1[4][6] =
{
{
{1.000000f, 0.995185f, 0.980785f, 0.956940f},
{0.000000f, -0.098017f, -0.195090f, -0.290285f},
{1.000000f, 0.980785f, 0.923880f, 0.831470f},
{0.000000f, -0.195090f, -0.382683f, -0.555570f},
{1.000000f, 0.956940f, 0.831470f, 0.634393f},
{0.000000f, -0.290285f, -0.555570f, -0.773010f},
},
{
{0.923880f, 0.881921f, 0.831470f, 0.773010f},
{-0.382683f, -0.471397f, -0.555570f, -0.634393f},
{0.707107f, 0.555570f, 0.382683f, 0.195090f},
{-0.707107f, -0.831470f, -0.923880f, -0.980785f},
{0.382683f, 0.098017f, -0.195090f, -0.471397f},
{-0.923880f, -0.995185f, -0.980785f, -0.881921f},
},
{
{0.707107f, 0.634393f, 0.555570f, 0.471397f},
{-0.707107f, -0.773010f, -0.831470f, -0.881921f},
{0.000000f, -0.195090f, -0.382683f, -0.555570f},
{-1.000000f, -0.980785f, -0.923880f, -0.831470f},
{-0.707107f, -0.881921f, -0.980785f, -0.995185f},
{-0.707107f, -0.471397f, -0.195090f, 0.098017f},
},
{
{0.382683f, 0.290285f, 0.195090f, 0.098017f},
{-0.923880f, -0.956940f, -0.980785f, -0.995185f},
{-0.707107f, -0.831470f, -0.923880f, -0.980785f},
{-0.707107f, -0.555570f, -0.382683f, -0.195090f},
{-0.923880f, -0.773010f, -0.555570f, -0.290285f},
{0.382683f, 0.634393f, 0.831470f, 0.956940f},
}
};
_declspec(align(16)) SConstSSE _tabCoef64_2[4][6] =
{
{
{1.000000f, 0.923880f, 0.707107f, 0.382683f},
{0.000000f, -0.382683f, -0.707107f, -0.923880f},
{1.000000f, 0.707107f, 0.000000f, -0.707107f},
{0.000000f, -0.707107f, -1.000000f, -0.707107f},
{1.000000f, 0.382683f, -0.707107f, -0.923880f},
{0.000000f, -0.923880f, -0.707107f, 0.382683f},
}
};
void cradix4c_64(float* ar, float* ai, int nm)
{
int wdt = OCEANGRID;
float *arw = &ar[wdt];
float *aiw = &ai[wdt];
_asm
{
mov eax, arw
mov ebx, aiw
mov edx, ai
mov ecx, ar
sub esp, 20h
mov [esp+0ch], ebp
mov [esp], ecx
mov [esp+4], edx
mov ebp, ecx
mov [esp+10h], eax
mov [esp+14h], ebx
mov [esp+18h], ebp
mov [esp+1ch], edx
lea ebx, [ebp+40h]
lea edi, _tabCoef64_1
mov [esp+8], ebx
_lAlign:
movaps xmm0,xmmword ptr [ebp]
movaps xmm4,xmm0
movaps xmm2,xmmword ptr [ebp+80h]
subps xmm0,xmm2
movaps xmm1,xmmword ptr [ebp+40h]
addps xmm4,xmm2
movaps xmm3,xmmword ptr [ebp+0C0h]
movaps xmm5,xmm1
movaps xmm2,xmmword ptr [edx]
addps xmm5,xmm3
jmp _lCicleAlign
align 4
_lStartAlign:
movaps xmmword ptr [edx+0B0h],xmm0
movaps xmm0,xmmword ptr [ebp]
subps xmm7,xmm4
movaps xmm4,xmm0
movaps xmmword ptr [ebp+0B0h],xmm5
addps xmm3,xmm1
movaps xmm2,xmmword ptr [ebp+80h]
movaps xmmword ptr [ebp+30h],xmm7
subps xmm0,xmm2
movaps xmm1,xmmword ptr [ebp+40h]
addps xmm4,xmm2
movaps xmm5,xmm1
movaps xmmword ptr [edx+30h],xmm3
movaps xmm2,xmmword ptr [edx]
movaps xmm3,xmmword ptr [ebp+0C0h]
addps xmm5,xmm3
_lCicleAlign:
align 4
movaps xmm7,xmm4
subps xmm1,xmm3
prefetcht0 [edi]
addps xmm4,xmm5
movaps xmm3,xmm2
movaps xmm6,xmmword ptr [edx+80h]
subps xmm7,xmm5
prefetcht0 [edi+10h]
movaps xmmword ptr [ebp],xmm4
addps xmm3,xmm6
movaps xmm5,xmmword ptr [edx+40h]
subps xmm2,xmm6
prefetcht0 [edi+20h]
movaps xmm4,xmmword ptr [edx+0C0h]
movaps xmm6,xmm5
addps xmm5,xmm4
prefetcht0 [edi+30h]
subps xmm6,xmm4
movaps xmm4,xmm5
addps xmm5,xmm3
movaps xmmword ptr [edx],xmm5
movaps xmm5,xmm0
addps xmm0,xmm6
prefetcht0 [edi+40h]
subps xmm3,xmm4
movaps xmm4,xmm2
subps xmm2,xmm1
prefetcht0 [edi+50h]
addps xmm1,xmm4
subps xmm5,xmm6
movaps xmm4,xmm0
movaps xmm6,xmm2
mulps xmm0,xmmword ptr [edi]
mulps xmm2,xmmword ptr [edi+10h]
mulps xmm4,xmmword ptr [edi+10h]
add ebp,10h
add edx,10h
subps xmm0,xmm2
mulps xmm6,xmmword ptr [edi]
movaps xmm2,xmm1
movaps xmmword ptr [ebp+70h],xmm0
mulps xmm1,xmmword ptr [edi+50h]
movaps xmm0,xmm5
addps xmm4,xmm6
mulps xmm5,xmmword ptr [edi+40h]
mulps xmm2,xmmword ptr [edi+40h]
mulps xmm0,xmmword ptr [edi+50h]
subps xmm5,xmm1
movaps xmmword ptr [edx+70h],xmm4
movaps xmm1,xmm7
mulps xmm7,xmmword ptr [edi+20h]
movaps xmm4,xmm3
addps xmm0,xmm2
mulps xmm3,xmmword ptr [edi+20h]
mulps xmm4,xmmword ptr [edi+30h]
cmp ebp,ebx
mulps xmm1,xmmword ptr [edi+30h]
lea edi,[edi+60h]
jl _lStartAlign
movaps xmmword ptr [ebp+0B0h],xmm5
addps xmm3,xmm1
subps xmm7,xmm4
movaps xmmword ptr [edx+0B0h],xmm0
movaps xmmword ptr [edx+30h],xmm3
movaps xmmword ptr [ebp+30h],xmm7
mov ebp,[esp]
mov edx,[esp+4]
add ebp,100h
add edx,100h
mov [esp],ebp
mov [esp+4],edx
cmp ebp,[esp+10h]
lea ebx,[ebp+40h]
mov [esp+8],ebx
lea edi,_tabCoef64_1
jl _lAlign
mov ebp, [esp+18h];
mov edx, [esp+1ch];
mov [esp], ebp
mov [esp+4], edx
lea ebx, [ebp+10h]
lea edi, _tabCoef64_2
mov [esp+8], ebx
_lAlign2:
movaps xmm0,xmmword ptr [ebp]
movaps xmm4,xmm0
movaps xmm2,xmmword ptr [ebp+20h]
subps xmm0,xmm2
movaps xmm1,xmmword ptr [ebp+10h]
addps xmm4,xmm2
movaps xmm3,xmmword ptr [ebp+30h]
movaps xmm5,xmm1
movaps xmm2,xmmword ptr [edx]
addps xmm5,xmm3
jmp _lCicleAlign2
align 4
_lStartAlign2:
movaps xmmword ptr [edx+20h],xmm0
movaps xmm0,xmmword ptr [ebp]
subps xmm7,xmm4
movaps xmm4,xmm0
movaps xmmword ptr [ebp+20h],xmm5
addps xmm3,xmm1
movaps xmm2,xmmword ptr [ebp+20h]
movaps xmmword ptr [ebp],xmm7
subps xmm0,xmm2
movaps xmm1,xmmword ptr [ebp+10h]
addps xmm4,xmm2
movaps xmm5,xmm1
movaps xmmword ptr [edx],xmm3
movaps xmm2,xmmword ptr [edx]
movaps xmm3,xmmword ptr [ebp+30h]
addps xmm5,xmm3
_lCicleAlign2:
movaps xmm7,xmm4
subps xmm1,xmm3
prefetcht0 [edi]
addps xmm4,xmm5
movaps xmm3,xmm2
movaps xmm6,xmmword ptr [edx+20h]
subps xmm7,xmm5
prefetcht0 [edi+10h]
movaps xmmword ptr [ebp],xmm4
addps xmm3,xmm6
movaps xmm5,xmmword ptr [edx+10h]
subps xmm2,xmm6
prefetcht0 [edi+20h]
movaps xmm4,xmmword ptr [edx+30h]
movaps xmm6,xmm5
addps xmm5,xmm4
prefetcht0 [edi+30h]
subps xmm6,xmm4
movaps xmm4,xmm5
addps xmm5,xmm3
movaps xmmword ptr [edx],xmm5
movaps xmm5,xmm0
addps xmm0,xmm6
prefetcht0 [edi+40h]
subps xmm3,xmm4
movaps xmm4,xmm2
subps xmm2,xmm1
prefetcht0 [edi+50h]
addps xmm1,xmm4
subps xmm5,xmm6
movaps xmm4,xmm0
movaps xmm6,xmm2
mulps xmm0,xmmword ptr [edi]
mulps xmm2,xmmword ptr [edi+10h]
mulps xmm4,xmmword ptr [edi+10h]
add ebp,10h
add edx,10h
subps xmm0,xmm2
mulps xmm6,xmmword ptr [edi]
movaps xmm2,xmm1
movaps xmmword ptr [ebp+10h],xmm0
mulps xmm1,xmmword ptr [edi+50h]
movaps xmm0,xmm5
addps xmm4,xmm6
mulps xmm5,xmmword ptr [edi+40h]
mulps xmm2,xmmword ptr [edi+40h]
mulps xmm0,xmmword ptr [edi+50h]
subps xmm5,xmm1
movaps xmmword ptr [edx+10h],xmm4
movaps xmm1,xmm7
mulps xmm7,xmmword ptr [edi+20h]
movaps xmm4,xmm3
addps xmm0,xmm2
mulps xmm3,xmmword ptr [edi+20h]
mulps xmm4,xmmword ptr [edi+30h]
cmp ebp,ebx
mulps xmm1,xmmword ptr [edi+30h]
lea edi,[edi+60h]
jl _lStartAlign2
movaps xmmword ptr [ebp+20h],xmm5
addps xmm3,xmm1
subps xmm7,xmm4
movaps xmmword ptr [edx+20h],xmm0
movaps xmmword ptr [edx],xmm3
movaps xmmword ptr [ebp],xmm7
mov ebp,[esp]
mov edx,[esp+4]
add ebp,40h
add edx,40h
mov [esp],ebp
mov [esp+4],edx
cmp ebp,[esp+10h]
lea ebx,[ebp+10h]
mov [esp+8],ebx
lea edi,_tabCoef64_2
jl _lAlign2
mov ebp, [esp+18h]
mov edx, [esp+1ch]
mov [esp], ebp
mov [esp+4], edx
mov ebx, [esp+10h]
fld dword ptr [ebp]
fld st(0)
fld dword ptr [ebp+8]
fsub st(1),st
faddp st(2),st
fld dword ptr [ebp+4]
fxch st(2)
fld st(2)
fld dword ptr [ebp+0Ch]
fadd st(1),st
fsubp st(4),st
jmp _lEnd3
align 4
_lCicle:
fstp dword ptr [ebp-4]
fld dword ptr [ebp]
fxch st(1)
fstp dword ptr [ebp-8]
fld st(0)
fld dword ptr [ebp+8]
fadd st(2),st
fxch st(3)
fstp dword ptr [edx-10h]
fsubrp st(2),st
fld dword ptr [ebp+4]
fld st(0)
fld dword ptr [ebp+0Ch]
fadd st(2),st
fxch st(5)
fstp dword ptr [edx-4]
fsubrp st(4),st
_lEnd3:
fld dword ptr [edx]
fld dword ptr [edx]
fld st(3)
fadd st,st(3)
fld dword ptr [edx+8]
fadd st(3),st
fxch st(5)
fsubrp st(4),st
fld dword ptr [edx+4]
fxch st(2)
fsubrp st(5),st
fstp dword ptr [ebp]
fld st(0)
fld dword ptr [edx+0Ch]
fadd st(2),st
fxch st(4)
fstp dword ptr [ebp+4]
fsubrp st(3),st
fst [esp+10h]
fadd st,st(1)
fxch st(4)
fst [esp+14h]
fadd st,st(2)
fxch st(3)
fst [esp+18h]
fsub st,st(5)
fld [esp+10h]
fsubp st(2),st
fld [esp+14h]
fsubrp st(3),st
fld [esp+18h]
faddp st(6),st
add ebp,10h
add edx,10h
fstp dword ptr [edx-8]
cmp ebp,ebx
fstp dword ptr [edx-0Ch]
jl _lCicle
fstp dword ptr [ebp-4]
fstp dword ptr [ebp-8]
fstp dword ptr [edx-10h]
fstp dword ptr [edx-4]
mov ebp, [esp+0ch]
add esp, 20h
}
}
void bittabc(int *p, int sn)
{
int i2 = sn;
int j = 1;
int i, k;
int ind = 0;
i2 >>= 1;
for (i=1; i<=sn-1; i++)
{
if (i < j)
{
ind += 2;
p[ind] = i-1;
p[ind+1] = j-1;
}
k = i2;
while (j > k)
{
j -= k;
k >>= 1;
}
j += k;
}
p[0] = 0;
p[1] = ind >> 1;
}
void coef4r22c(int *ptr, int nm)
{
int sn = 1<<nm;
bittabc(ptr, sn);
}
void cbitrevc(float* ar, float* ai, int *p)
{
int nCount = p[1];
int *pInd = &p[2];
int nC = nCount >> 2;
while (nC)
{
Exchange(ar[pInd[0]], ar[pInd[1]]);
Exchange(ar[pInd[2]], ar[pInd[3]]);
Exchange(ar[pInd[4]], ar[pInd[5]]);
Exchange(ar[pInd[6]], ar[pInd[7]]);
pInd += 8;
nC--;
}
pInd = &p[2];
nC = nCount >> 2;
while (nC)
{
Exchange(ai[pInd[0]], ai[pInd[1]]);
Exchange(ai[pInd[2]], ai[pInd[3]]);
Exchange(ai[pInd[4]], ai[pInd[5]]);
Exchange(ai[pInd[6]], ai[pInd[7]]);
pInd += 8;
nC--;
}
}
void xcfft1dc(float* ar, float* ai, int *p, int nm)
{
cradix4c_64(ar, ai, nm);
cbitrevc(ar, ai, p);
}
void FFTSSE_64(float* ar, float* ai)
{
int i, j;
const int nm = 6;
_declspec(align(16)) int p0[OCEANGRID*8];
coef4r22c(p0, nm);
for (i=0; i<OCEANGRID; i++)
{
xcfft1dc(&ar[OCEANGRID*i], &ai[OCEANGRID*i], p0, nm);
}
coef4r22c(p0, nm);
_declspec(align(16)) float p1[4][OCEANGRID];
_declspec(align(16)) float p2[4][OCEANGRID];
float *src, *dst;
for (j=0; j<OCEANGRID; j+=4)
{
src = &ar[j];
for (i=0; i<OCEANGRID; i++, src+=OCEANGRID)
{
p1[0][i] = src[0];
p1[1][i] = src[1];
p1[2][i] = src[2];
p1[3][i] = src[3];
}
src = &ai[j];
for (i=0; i<OCEANGRID; i++, src+=OCEANGRID)
{
p2[0][i] = src[0];
p2[1][i] = src[1];
p2[2][i] = src[2];
p2[3][i] = src[3];
}
xcfft1dc(&p1[0][0], &p2[0][0], p0, nm);
xcfft1dc(&p1[1][0], &p2[1][0], p0, nm);
xcfft1dc(&p1[2][0], &p2[2][0], p0, nm);
xcfft1dc(&p1[3][0], &p2[3][0], p0, nm);
dst = &ar[j];
for (i=0; i<OCEANGRID; i++, dst+=OCEANGRID)
{
dst[0] = p1[0][i];
dst[1] = p1[1][i];
dst[2] = p1[2][i];
dst[3] = p1[3][i];
}
dst = &ai[j];
for (i=0; i<OCEANGRID; i++, dst+=OCEANGRID)
{
dst[0] = p2[0][i];
dst[1] = p2[1][i];
dst[2] = p2[2][i];
dst[3] = p2[3][i];
}
}
}
#pragma warning(pop)
#endif // !defined(WIN64)