quda-ref/v0.7.0/tm__dslash__dagger__g80__core_8h_source.html

 // *** CUDA DSLASH DAGGER ***


 #define DSLASH_SHARED_FLOATS_PER_THREAD 19


 #if ((CUDA_VERSION >= 4010) && (__COMPUTE_CAPABILITY__ >= 200)) // NVVM compiler

 #define VOLATILE

 #else // Open64 compiler

 #define VOLATILE volatile

 #endif

 // input spinor

 #ifdef SPINOR_DOUBLE

 #define spinorFloat double

 #define i00_re I0.x

 #define i00_im I0.y

 #define i01_re I1.x

 #define i01_im I1.y

 #define i02_re I2.x

 #define i02_im I2.y

 #define i10_re I3.x

 #define i10_im I3.y

 #define i11_re I4.x

 #define i11_im I4.y

 #define i12_re I5.x

 #define i12_im I5.y

 #define i20_re I6.x

 #define i20_im I6.y

 #define i21_re I7.x

 #define i21_im I7.y

 #define i22_re I8.x

 #define i22_im I8.y

 #define i30_re I9.x

 #define i30_im I9.y

 #define i31_re I10.x

 #define i31_im I10.y

 #define i32_re I11.x

 #define i32_im I11.y

 #define acc00_re accum0.x

 #define acc00_im accum0.y

 #define acc01_re accum1.x

 #define acc01_im accum1.y

 #define acc02_re accum2.x

 #define acc02_im accum2.y

 #define acc10_re accum3.x

 #define acc10_im accum3.y

 #define acc11_re accum4.x

 #define acc11_im accum4.y

 #define acc12_re accum5.x

 #define acc12_im accum5.y

 #define acc20_re accum6.x

 #define acc20_im accum6.y

 #define acc21_re accum7.x

 #define acc21_im accum7.y

 #define acc22_re accum8.x

 #define acc22_im accum8.y

 #define acc30_re accum9.x

 #define acc30_im accum9.y

 #define acc31_re accum10.x

 #define acc31_im accum10.y

 #define acc32_re accum11.x

 #define acc32_im accum11.y

 #else

 #define spinorFloat float

 #define i00_re I0.x

 #define i00_im I0.y

 #define i01_re I0.z

 #define i01_im I0.w

 #define i02_re I1.x

 #define i02_im I1.y

 #define i10_re I1.z

 #define i10_im I1.w

 #define i11_re I2.x

 #define i11_im I2.y

 #define i12_re I2.z

 #define i12_im I2.w

 #define i20_re I3.x

 #define i20_im I3.y

 #define i21_re I3.z

 #define i21_im I3.w

 #define i22_re I4.x

 #define i22_im I4.y

 #define i30_re I4.z

 #define i30_im I4.w

 #define i31_re I5.x

 #define i31_im I5.y

 #define i32_re I5.z

 #define i32_im I5.w

 #define acc00_re accum0.x

 #define acc00_im accum0.y

 #define acc01_re accum0.z

 #define acc01_im accum0.w

 #define acc02_re accum1.x

 #define acc02_im accum1.y

 #define acc10_re accum1.z

 #define acc10_im accum1.w

 #define acc11_re accum2.x

 #define acc11_im accum2.y

 #define acc12_re accum2.z

 #define acc12_im accum2.w

 #define acc20_re accum3.x

 #define acc20_im accum3.y

 #define acc21_re accum3.z

 #define acc21_im accum3.w

 #define acc22_re accum4.x

 #define acc22_im accum4.y

 #define acc30_re accum4.z

 #define acc30_im accum4.w

 #define acc31_re accum5.x

 #define acc31_im accum5.y

 #define acc32_re accum5.z

 #define acc32_im accum5.w

 #endif // SPINOR_DOUBLE


 // gauge link

 #ifdef GAUGE_FLOAT2

 #define g00_re G0.x

 #define g00_im G0.y

 #define g01_re G1.x

 #define g01_im G1.y

 #define g02_re G2.x

 #define g02_im G2.y

 #define g10_re G3.x

 #define g10_im G3.y

 #define g11_re G4.x

 #define g11_im G4.y

 #define g12_re G5.x

 #define g12_im G5.y

 #define g20_re G6.x

 #define g20_im G6.y

 #define g21_re G7.x

 #define g21_im G7.y

 #define g22_re G8.x

 #define g22_im G8.y


 #else

 #define g00_re G0.x

 #define g00_im G0.y

 #define g01_re G0.z

 #define g01_im G0.w

 #define g02_re G1.x

 #define g02_im G1.y

 #define g10_re G1.z

 #define g10_im G1.w

 #define g11_re G2.x

 #define g11_im G2.y

 #define g12_re G2.z

 #define g12_im G2.w

 #define g20_re G3.x

 #define g20_im G3.y

 #define g21_re G3.z

 #define g21_im G3.w

 #define g22_re G4.x

 #define g22_im G4.y


 #endif // GAUGE_DOUBLE


 // conjugated gauge link

 #define gT00_re (+g00_re)

 #define gT00_im (-g00_im)

 #define gT01_re (+g10_re)

 #define gT01_im (-g10_im)

 #define gT02_re (+g20_re)

 #define gT02_im (-g20_im)

 #define gT10_re (+g01_re)

 #define gT10_im (-g01_im)

 #define gT11_re (+g11_re)

 #define gT11_im (-g11_im)

 #define gT12_re (+g21_re)

 #define gT12_im (-g21_im)

 #define gT20_re (+g02_re)

 #define gT20_im (-g02_im)

 #define gT21_re (+g12_re)

 #define gT21_im (-g12_im)

 #define gT22_re (+g22_re)

 #define gT22_im (-g22_im)


 // output spinor

 #define o00_re s[0*SHARED_STRIDE]

 #define o00_im s[1*SHARED_STRIDE]

 #define o01_re s[2*SHARED_STRIDE]

 #define o01_im s[3*SHARED_STRIDE]

 #define o02_re s[4*SHARED_STRIDE]

 #define o02_im s[5*SHARED_STRIDE]

 #define o10_re s[6*SHARED_STRIDE]

 #define o10_im s[7*SHARED_STRIDE]

 #define o11_re s[8*SHARED_STRIDE]

 #define o11_im s[9*SHARED_STRIDE]

 #define o12_re s[10*SHARED_STRIDE]

 #define o12_im s[11*SHARED_STRIDE]

 #define o20_re s[12*SHARED_STRIDE]

 #define o20_im s[13*SHARED_STRIDE]

 #define o21_re s[14*SHARED_STRIDE]

 #define o21_im s[15*SHARED_STRIDE]

 #define o22_re s[16*SHARED_STRIDE]

 #define o22_im s[17*SHARED_STRIDE]

 #define o30_re s[18*SHARED_STRIDE]

 VOLATILE spinorFloat o30_im;

 VOLATILE spinorFloat o31_re;

 VOLATILE spinorFloat o31_im;

 VOLATILE spinorFloat o32_re;

 VOLATILE spinorFloat o32_im;


 #ifdef SPINOR_DOUBLE

 #define SHARED_STRIDE  8 // to avoid bank conflicts on G80 and GT200

 #else

 #define SHARED_STRIDE 16 // to avoid bank conflicts on G80 and GT200

 #endif


 extern __shared__ char s_data[];


 VOLATILE spinorFloat *s = (spinorFloat*)s_data + DSLASH_SHARED_FLOATS_PER_THREAD*SHARED_STRIDE*(threadIdx.x/SHARED_STRIDE)

                                   + (threadIdx.x % SHARED_STRIDE);


 #include "read_gauge.h"

 #include "io_spinor.h"


 int x1, x2, x3, x4;

 int X;


 #if (defined MULTI_GPU) && (DD_PREC==2) // half precision

 int sp_norm_idx;

 #endif // MULTI_GPU half precision


 int sid;


 #ifdef MULTI_GPU

 int face_idx;

 if (kernel_type == INTERIOR_KERNEL) {

 #endif


   sid = blockIdx.x*blockDim.x + threadIdx.x;

   if (sid >= param.threads) return;


   // Inline by hand for the moment and assume even dimensions

   const int dims[] = {X1, X2, X3, X4};

   coordsFromIndex<EVEN_X>(X, x1, x2, x3, x4, sid, param.parity, dims);


   o00_re = 0;  o00_im = 0;

   o01_re = 0;  o01_im = 0;

   o02_re = 0;  o02_im = 0;

   o10_re = 0;  o10_im = 0;

   o11_re = 0;  o11_im = 0;

   o12_re = 0;  o12_im = 0;

   o20_re = 0;  o20_im = 0;

   o21_re = 0;  o21_im = 0;

   o22_re = 0;  o22_im = 0;

   o30_re = 0;  o30_im = 0;

   o31_re = 0;  o31_im = 0;

   o32_re = 0;  o32_im = 0;


 #ifdef MULTI_GPU

 } else { // exterior kernel


   sid = blockIdx.x*blockDim.x + threadIdx.x;

   if (sid >= param.threads) return;


   const int dim = static_cast<int>(kernel_type);

   const int face_volume = (param.threads >> 1);           // volume of one face

   const int face_num = (sid >= face_volume);              // is this thread updating face 0 or 1

   face_idx = sid - face_num*face_volume;        // index into the respective face


   // ghostOffset is scaled to include body (includes stride) and number of FloatN arrays (SPINOR_HOP)

   // face_idx not sid since faces are spin projected and share the same volume index (modulo UP/DOWN reading)

   //sp_idx = face_idx + param.ghostOffset[dim];


 #if (DD_PREC==2) // half precision

   sp_norm_idx = sid + param.ghostNormOffset[static_cast<int>(kernel_type)];

 #endif


   const int dims[] = {X1, X2, X3, X4};

   coordsFromFaceIndex<1>(X, sid, x1, x2, x3, x4, face_idx, face_volume, dim, face_num, param.parity, dims);


   READ_INTERMEDIATE_SPINOR(INTERTEX, param.sp_stride, sid, sid);


   o00_re = i00_re;  o00_im = i00_im;

   o01_re = i01_re;  o01_im = i01_im;

   o02_re = i02_re;  o02_im = i02_im;

   o10_re = i10_re;  o10_im = i10_im;

   o11_re = i11_re;  o11_im = i11_im;

   o12_re = i12_re;  o12_im = i12_im;

   o20_re = i20_re;  o20_im = i20_im;

   o21_re = i21_re;  o21_im = i21_im;

   o22_re = i22_re;  o22_im = i22_im;

   o30_re = i30_re;  o30_im = i30_im;

   o31_re = i31_re;  o31_im = i31_im;

   o32_re = i32_re;  o32_im = i32_im;

 }

 #endif // MULTI_GPU


 #ifdef MULTI_GPU

 if ( (kernel_type == INTERIOR_KERNEL && (!param.ghostDim[0] || x1<X1m1)) ||

      (kernel_type == EXTERIOR_KERNEL_X && x1==X1m1) )

 #endif

 {

   // Projector P0+

   // 1 0 0 i

   // 0 1 i 0

   // 0 -i 1 0

   // -i 0 0 1


 #ifdef MULTI_GPU

   const int sp_idx = (kernel_type == INTERIOR_KERNEL) ? (x1==X1m1 ? X-X1m1 : X+1) >> 1 :

     face_idx + param.ghostOffset[static_cast<int>(kernel_type)];

 #else

   const int sp_idx = (x1==X1m1 ? X-X1m1 : X+1) >> 1;

 #endif


   const int ga_idx = sid;


   spinorFloat a0_re, a0_im;

   spinorFloat a1_re, a1_im;

   spinorFloat a2_re, a2_im;

   spinorFloat b0_re, b0_im;

   spinorFloat b1_re, b1_im;

   spinorFloat b2_re, b2_im;


 #ifdef MULTI_GPU

   if (kernel_type == INTERIOR_KERNEL) {

 #endif


     // read spinor from device memory

     READ_SPINOR(SPINORTEX, param.sp_stride, sp_idx, sp_idx);

 #ifdef TWIST_INV_DSLASH

     APPLY_TWIST_INV(-a, b, i);

 #endif


     // project spinor into half spinors

     a0_re = +i00_re-i30_im;

     a0_im = +i00_im+i30_re;

     a1_re = +i01_re-i31_im;

     a1_im = +i01_im+i31_re;

     a2_re = +i02_re-i32_im;

     a2_im = +i02_im+i32_re;

     b0_re = +i10_re-i20_im;

     b0_im = +i10_im+i20_re;

     b1_re = +i11_re-i21_im;

     b1_im = +i11_im+i21_re;

     b2_re = +i12_re-i22_im;

     b2_im = +i12_im+i22_re;


 #ifdef MULTI_GPU

   } else {


     const int sp_stride_pad = ghostFace[static_cast<int>(kernel_type)];


     // read half spinor from device memory

     READ_HALF_SPINOR(SPINORTEX, sp_stride_pad, sp_idx + (SPINOR_HOP/2)*sp_stride_pad, sp_norm_idx);


     a0_re = i00_re;  a0_im = i00_im;

     a1_re = i01_re;  a1_im = i01_im;

     a2_re = i02_re;  a2_im = i02_im;

     b0_re = i10_re;  b0_im = i10_im;

     b1_re = i11_re;  b1_im = i11_im;

     b2_re = i12_re;  b2_im = i12_im;


   }

 #endif // MULTI_GPU


   // read gauge matrix from device memory

   READ_GAUGE_MATRIX(G, GAUGE0TEX, 0, ga_idx, ga_stride);


   // reconstruct gauge matrix

   RECONSTRUCT_GAUGE_MATRIX(0);


   // multiply row 0

   spinorFloat A0_re = 0;

   A0_re += g00_re * a0_re;

   A0_re -= g00_im * a0_im;

   A0_re += g01_re * a1_re;

   A0_re -= g01_im * a1_im;

   A0_re += g02_re * a2_re;

   A0_re -= g02_im * a2_im;

   spinorFloat A0_im = 0;

   A0_im += g00_re * a0_im;

   A0_im += g00_im * a0_re;

   A0_im += g01_re * a1_im;

   A0_im += g01_im * a1_re;

   A0_im += g02_re * a2_im;

   A0_im += g02_im * a2_re;

   spinorFloat B0_re = 0;

   B0_re += g00_re * b0_re;

   B0_re -= g00_im * b0_im;

   B0_re += g01_re * b1_re;

   B0_re -= g01_im * b1_im;

   B0_re += g02_re * b2_re;

   B0_re -= g02_im * b2_im;

   spinorFloat B0_im = 0;

   B0_im += g00_re * b0_im;

   B0_im += g00_im * b0_re;

   B0_im += g01_re * b1_im;

   B0_im += g01_im * b1_re;

   B0_im += g02_re * b2_im;

   B0_im += g02_im * b2_re;


   // multiply row 1

   spinorFloat A1_re = 0;

   A1_re += g10_re * a0_re;

   A1_re -= g10_im * a0_im;

   A1_re += g11_re * a1_re;

   A1_re -= g11_im * a1_im;

   A1_re += g12_re * a2_re;

   A1_re -= g12_im * a2_im;

   spinorFloat A1_im = 0;

   A1_im += g10_re * a0_im;

   A1_im += g10_im * a0_re;

   A1_im += g11_re * a1_im;

   A1_im += g11_im * a1_re;

   A1_im += g12_re * a2_im;

   A1_im += g12_im * a2_re;

   spinorFloat B1_re = 0;

   B1_re += g10_re * b0_re;

   B1_re -= g10_im * b0_im;

   B1_re += g11_re * b1_re;

   B1_re -= g11_im * b1_im;

   B1_re += g12_re * b2_re;

   B1_re -= g12_im * b2_im;

   spinorFloat B1_im = 0;

   B1_im += g10_re * b0_im;

   B1_im += g10_im * b0_re;

   B1_im += g11_re * b1_im;

   B1_im += g11_im * b1_re;

   B1_im += g12_re * b2_im;

   B1_im += g12_im * b2_re;


   // multiply row 2

   spinorFloat A2_re = 0;

   A2_re += g20_re * a0_re;

   A2_re -= g20_im * a0_im;

   A2_re += g21_re * a1_re;

   A2_re -= g21_im * a1_im;

   A2_re += g22_re * a2_re;

   A2_re -= g22_im * a2_im;

   spinorFloat A2_im = 0;

   A2_im += g20_re * a0_im;

   A2_im += g20_im * a0_re;

   A2_im += g21_re * a1_im;

   A2_im += g21_im * a1_re;

   A2_im += g22_re * a2_im;

   A2_im += g22_im * a2_re;

   spinorFloat B2_re = 0;

   B2_re += g20_re * b0_re;

   B2_re -= g20_im * b0_im;

   B2_re += g21_re * b1_re;

   B2_re -= g21_im * b1_im;

   B2_re += g22_re * b2_re;

   B2_re -= g22_im * b2_im;

   spinorFloat B2_im = 0;

   B2_im += g20_re * b0_im;

   B2_im += g20_im * b0_re;

   B2_im += g21_re * b1_im;

   B2_im += g21_im * b1_re;

   B2_im += g22_re * b2_im;

   B2_im += g22_im * b2_re;


   o00_re += A0_re;

   o00_im += A0_im;

   o10_re += B0_re;

   o10_im += B0_im;

   o20_re += B0_im;

   o20_im -= B0_re;

   o30_re += A0_im;

   o30_im -= A0_re;


   o01_re += A1_re;

   o01_im += A1_im;

   o11_re += B1_re;

   o11_im += B1_im;

   o21_re += B1_im;

   o21_im -= B1_re;

   o31_re += A1_im;

   o31_im -= A1_re;


   o02_re += A2_re;

   o02_im += A2_im;

   o12_re += B2_re;

   o12_im += B2_im;

   o22_re += B2_im;

   o22_im -= B2_re;

   o32_re += A2_im;

   o32_im -= A2_re;


 }


 #ifdef MULTI_GPU

 if ( (kernel_type == INTERIOR_KERNEL && (!param.ghostDim[0] || x1>0)) ||

      (kernel_type == EXTERIOR_KERNEL_X && x1==0) )

 #endif

 {

   // Projector P0-

   // 1 0 0 -i

   // 0 1 -i 0

   // 0 i 1 0

   // i 0 0 1


 #ifdef MULTI_GPU

   const int sp_idx = (kernel_type == INTERIOR_KERNEL) ? (x1==0 ? X+X1m1 : X-1) >> 1 :

     face_idx + param.ghostOffset[static_cast<int>(kernel_type)];

 #else

   const int sp_idx = (x1==0 ? X+X1m1 : X-1) >> 1;

 #endif


 #ifdef MULTI_GPU

   const int ga_idx = ((kernel_type == INTERIOR_KERNEL) ? sp_idx : Vh+face_idx);

 #else

   const int ga_idx = sp_idx;

 #endif


   spinorFloat a0_re, a0_im;

   spinorFloat a1_re, a1_im;

   spinorFloat a2_re, a2_im;

   spinorFloat b0_re, b0_im;

   spinorFloat b1_re, b1_im;

   spinorFloat b2_re, b2_im;


 #ifdef MULTI_GPU

   if (kernel_type == INTERIOR_KERNEL) {

 #endif


     // read spinor from device memory

     READ_SPINOR(SPINORTEX, param.sp_stride, sp_idx, sp_idx);

 #ifdef TWIST_INV_DSLASH

     APPLY_TWIST_INV(-a, b, i);

 #endif


     // project spinor into half spinors

     a0_re = +i00_re+i30_im;

     a0_im = +i00_im-i30_re;

     a1_re = +i01_re+i31_im;

     a1_im = +i01_im-i31_re;

     a2_re = +i02_re+i32_im;

     a2_im = +i02_im-i32_re;

     b0_re = +i10_re+i20_im;

     b0_im = +i10_im-i20_re;

     b1_re = +i11_re+i21_im;

     b1_im = +i11_im-i21_re;

     b2_re = +i12_re+i22_im;

     b2_im = +i12_im-i22_re;


 #ifdef MULTI_GPU

   } else {


     const int sp_stride_pad = ghostFace[static_cast<int>(kernel_type)];


     // read half spinor from device memory

     READ_HALF_SPINOR(SPINORTEX, sp_stride_pad, sp_idx, sp_norm_idx);


     a0_re = i00_re;  a0_im = i00_im;

     a1_re = i01_re;  a1_im = i01_im;

     a2_re = i02_re;  a2_im = i02_im;

     b0_re = i10_re;  b0_im = i10_im;

     b1_re = i11_re;  b1_im = i11_im;

     b2_re = i12_re;  b2_im = i12_im;


   }

 #endif // MULTI_GPU


   // read gauge matrix from device memory

   READ_GAUGE_MATRIX(G, GAUGE1TEX, 1, ga_idx, ga_stride);


   // reconstruct gauge matrix

   RECONSTRUCT_GAUGE_MATRIX(1);


   // multiply row 0

   spinorFloat A0_re = 0;

   A0_re += gT00_re * a0_re;

   A0_re -= gT00_im * a0_im;

   A0_re += gT01_re * a1_re;

   A0_re -= gT01_im * a1_im;

   A0_re += gT02_re * a2_re;

   A0_re -= gT02_im * a2_im;

   spinorFloat A0_im = 0;

   A0_im += gT00_re * a0_im;

   A0_im += gT00_im * a0_re;

   A0_im += gT01_re * a1_im;

   A0_im += gT01_im * a1_re;

   A0_im += gT02_re * a2_im;

   A0_im += gT02_im * a2_re;

   spinorFloat B0_re = 0;

   B0_re += gT00_re * b0_re;

   B0_re -= gT00_im * b0_im;

   B0_re += gT01_re * b1_re;

   B0_re -= gT01_im * b1_im;

   B0_re += gT02_re * b2_re;

   B0_re -= gT02_im * b2_im;

   spinorFloat B0_im = 0;

   B0_im += gT00_re * b0_im;

   B0_im += gT00_im * b0_re;

   B0_im += gT01_re * b1_im;

   B0_im += gT01_im * b1_re;

   B0_im += gT02_re * b2_im;

   B0_im += gT02_im * b2_re;


   // multiply row 1

   spinorFloat A1_re = 0;

   A1_re += gT10_re * a0_re;

   A1_re -= gT10_im * a0_im;

   A1_re += gT11_re * a1_re;

   A1_re -= gT11_im * a1_im;

   A1_re += gT12_re * a2_re;

   A1_re -= gT12_im * a2_im;

   spinorFloat A1_im = 0;

   A1_im += gT10_re * a0_im;

   A1_im += gT10_im * a0_re;

   A1_im += gT11_re * a1_im;

   A1_im += gT11_im * a1_re;

   A1_im += gT12_re * a2_im;

   A1_im += gT12_im * a2_re;

   spinorFloat B1_re = 0;

   B1_re += gT10_re * b0_re;

   B1_re -= gT10_im * b0_im;

   B1_re += gT11_re * b1_re;

   B1_re -= gT11_im * b1_im;

   B1_re += gT12_re * b2_re;

   B1_re -= gT12_im * b2_im;

   spinorFloat B1_im = 0;

   B1_im += gT10_re * b0_im;

   B1_im += gT10_im * b0_re;

   B1_im += gT11_re * b1_im;

   B1_im += gT11_im * b1_re;

   B1_im += gT12_re * b2_im;

   B1_im += gT12_im * b2_re;


   // multiply row 2

   spinorFloat A2_re = 0;

   A2_re += gT20_re * a0_re;

   A2_re -= gT20_im * a0_im;

   A2_re += gT21_re * a1_re;

   A2_re -= gT21_im * a1_im;

   A2_re += gT22_re * a2_re;

   A2_re -= gT22_im * a2_im;

   spinorFloat A2_im = 0;

   A2_im += gT20_re * a0_im;

   A2_im += gT20_im * a0_re;

   A2_im += gT21_re * a1_im;

   A2_im += gT21_im * a1_re;

   A2_im += gT22_re * a2_im;

   A2_im += gT22_im * a2_re;

   spinorFloat B2_re = 0;

   B2_re += gT20_re * b0_re;

   B2_re -= gT20_im * b0_im;

   B2_re += gT21_re * b1_re;

   B2_re -= gT21_im * b1_im;

   B2_re += gT22_re * b2_re;

   B2_re -= gT22_im * b2_im;

   spinorFloat B2_im = 0;

   B2_im += gT20_re * b0_im;

   B2_im += gT20_im * b0_re;

   B2_im += gT21_re * b1_im;

   B2_im += gT21_im * b1_re;

   B2_im += gT22_re * b2_im;

   B2_im += gT22_im * b2_re;


   o00_re += A0_re;

   o00_im += A0_im;

   o10_re += B0_re;

   o10_im += B0_im;

   o20_re -= B0_im;

   o20_im += B0_re;

   o30_re -= A0_im;

   o30_im += A0_re;


   o01_re += A1_re;

   o01_im += A1_im;

   o11_re += B1_re;

   o11_im += B1_im;

   o21_re -= B1_im;

   o21_im += B1_re;

   o31_re -= A1_im;

   o31_im += A1_re;


   o02_re += A2_re;

   o02_im += A2_im;

   o12_re += B2_re;

   o12_im += B2_im;

   o22_re -= B2_im;

   o22_im += B2_re;

   o32_re -= A2_im;

   o32_im += A2_re;


 }


 #ifdef MULTI_GPU

 if ( (kernel_type == INTERIOR_KERNEL && (!param.ghostDim[1] || x2<X2m1)) ||

      (kernel_type == EXTERIOR_KERNEL_Y && x2==X2m1) )

 #endif

 {

   // Projector P1+

   // 1 0 0 1

   // 0 1 -1 0

   // 0 -1 1 0

   // 1 0 0 1


 #ifdef MULTI_GPU

   const int sp_idx = (kernel_type == INTERIOR_KERNEL) ? (x2==X2m1 ? X-X2X1mX1 : X+X1) >> 1 :

     face_idx + param.ghostOffset[static_cast<int>(kernel_type)];

 #else

   const int sp_idx = (x2==X2m1 ? X-X2X1mX1 : X+X1) >> 1;

 #endif


   const int ga_idx = sid;


   spinorFloat a0_re, a0_im;

   spinorFloat a1_re, a1_im;

   spinorFloat a2_re, a2_im;

   spinorFloat b0_re, b0_im;

   spinorFloat b1_re, b1_im;

   spinorFloat b2_re, b2_im;


 #ifdef MULTI_GPU

   if (kernel_type == INTERIOR_KERNEL) {

 #endif


     // read spinor from device memory

     READ_SPINOR(SPINORTEX, param.sp_stride, sp_idx, sp_idx);

 #ifdef TWIST_INV_DSLASH

     APPLY_TWIST_INV(-a, b, i);

 #endif


     // project spinor into half spinors

     a0_re = +i00_re+i30_re;

     a0_im = +i00_im+i30_im;

     a1_re = +i01_re+i31_re;

     a1_im = +i01_im+i31_im;

     a2_re = +i02_re+i32_re;

     a2_im = +i02_im+i32_im;

     b0_re = +i10_re-i20_re;

     b0_im = +i10_im-i20_im;

     b1_re = +i11_re-i21_re;

     b1_im = +i11_im-i21_im;

     b2_re = +i12_re-i22_re;

     b2_im = +i12_im-i22_im;


 #ifdef MULTI_GPU

   } else {


     const int sp_stride_pad = ghostFace[static_cast<int>(kernel_type)];


     // read half spinor from device memory

     READ_HALF_SPINOR(SPINORTEX, sp_stride_pad, sp_idx + (SPINOR_HOP/2)*sp_stride_pad, sp_norm_idx);


     a0_re = i00_re;  a0_im = i00_im;

     a1_re = i01_re;  a1_im = i01_im;

     a2_re = i02_re;  a2_im = i02_im;

     b0_re = i10_re;  b0_im = i10_im;

     b1_re = i11_re;  b1_im = i11_im;

     b2_re = i12_re;  b2_im = i12_im;


   }

 #endif // MULTI_GPU


   // read gauge matrix from device memory

   READ_GAUGE_MATRIX(G, GAUGE0TEX, 2, ga_idx, ga_stride);


   // reconstruct gauge matrix

   RECONSTRUCT_GAUGE_MATRIX(2);


   // multiply row 0

   spinorFloat A0_re = 0;

   A0_re += g00_re * a0_re;

   A0_re -= g00_im * a0_im;

   A0_re += g01_re * a1_re;

   A0_re -= g01_im * a1_im;

   A0_re += g02_re * a2_re;

   A0_re -= g02_im * a2_im;

   spinorFloat A0_im = 0;

   A0_im += g00_re * a0_im;

   A0_im += g00_im * a0_re;

   A0_im += g01_re * a1_im;

   A0_im += g01_im * a1_re;

   A0_im += g02_re * a2_im;

   A0_im += g02_im * a2_re;

   spinorFloat B0_re = 0;

   B0_re += g00_re * b0_re;

   B0_re -= g00_im * b0_im;

   B0_re += g01_re * b1_re;

   B0_re -= g01_im * b1_im;

   B0_re += g02_re * b2_re;

   B0_re -= g02_im * b2_im;

   spinorFloat B0_im = 0;

   B0_im += g00_re * b0_im;

   B0_im += g00_im * b0_re;

   B0_im += g01_re * b1_im;

   B0_im += g01_im * b1_re;

   B0_im += g02_re * b2_im;

   B0_im += g02_im * b2_re;


   // multiply row 1

   spinorFloat A1_re = 0;

   A1_re += g10_re * a0_re;

   A1_re -= g10_im * a0_im;

   A1_re += g11_re * a1_re;

   A1_re -= g11_im * a1_im;

   A1_re += g12_re * a2_re;

   A1_re -= g12_im * a2_im;

   spinorFloat A1_im = 0;

   A1_im += g10_re * a0_im;

   A1_im += g10_im * a0_re;

   A1_im += g11_re * a1_im;

   A1_im += g11_im * a1_re;

   A1_im += g12_re * a2_im;

   A1_im += g12_im * a2_re;

   spinorFloat B1_re = 0;

   B1_re += g10_re * b0_re;

   B1_re -= g10_im * b0_im;

   B1_re += g11_re * b1_re;

   B1_re -= g11_im * b1_im;

   B1_re += g12_re * b2_re;

   B1_re -= g12_im * b2_im;

   spinorFloat B1_im = 0;

   B1_im += g10_re * b0_im;

   B1_im += g10_im * b0_re;

   B1_im += g11_re * b1_im;

   B1_im += g11_im * b1_re;

   B1_im += g12_re * b2_im;

   B1_im += g12_im * b2_re;


   // multiply row 2

   spinorFloat A2_re = 0;

   A2_re += g20_re * a0_re;

   A2_re -= g20_im * a0_im;

   A2_re += g21_re * a1_re;

   A2_re -= g21_im * a1_im;

   A2_re += g22_re * a2_re;

   A2_re -= g22_im * a2_im;

   spinorFloat A2_im = 0;

   A2_im += g20_re * a0_im;

   A2_im += g20_im * a0_re;

   A2_im += g21_re * a1_im;

   A2_im += g21_im * a1_re;

   A2_im += g22_re * a2_im;

   A2_im += g22_im * a2_re;

   spinorFloat B2_re = 0;

   B2_re += g20_re * b0_re;

   B2_re -= g20_im * b0_im;

   B2_re += g21_re * b1_re;

   B2_re -= g21_im * b1_im;

   B2_re += g22_re * b2_re;

   B2_re -= g22_im * b2_im;

   spinorFloat B2_im = 0;

   B2_im += g20_re * b0_im;

   B2_im += g20_im * b0_re;

   B2_im += g21_re * b1_im;

   B2_im += g21_im * b1_re;

   B2_im += g22_re * b2_im;

   B2_im += g22_im * b2_re;


   o00_re += A0_re;

   o00_im += A0_im;

   o10_re += B0_re;

   o10_im += B0_im;

   o20_re -= B0_re;

   o20_im -= B0_im;

   o30_re += A0_re;

   o30_im += A0_im;


   o01_re += A1_re;

   o01_im += A1_im;

   o11_re += B1_re;

   o11_im += B1_im;

   o21_re -= B1_re;

   o21_im -= B1_im;

   o31_re += A1_re;

   o31_im += A1_im;


   o02_re += A2_re;

   o02_im += A2_im;

   o12_re += B2_re;

   o12_im += B2_im;

   o22_re -= B2_re;

   o22_im -= B2_im;

   o32_re += A2_re;

   o32_im += A2_im;


 }


 #ifdef MULTI_GPU

 if ( (kernel_type == INTERIOR_KERNEL && (!param.ghostDim[1] || x2>0)) ||

      (kernel_type == EXTERIOR_KERNEL_Y && x2==0) )

 #endif

 {

   // Projector P1-

   // 1 0 0 -1

   // 0 1 1 0

   // 0 1 1 0

   // -1 0 0 1


 #ifdef MULTI_GPU

   const int sp_idx = (kernel_type == INTERIOR_KERNEL) ? (x2==0 ? X+X2X1mX1 : X-X1) >> 1 :

     face_idx + param.ghostOffset[static_cast<int>(kernel_type)];

 #else

   const int sp_idx = (x2==0 ? X+X2X1mX1 : X-X1) >> 1;

 #endif


 #ifdef MULTI_GPU

   const int ga_idx = ((kernel_type == INTERIOR_KERNEL) ? sp_idx : Vh+face_idx);

 #else

   const int ga_idx = sp_idx;

 #endif


   spinorFloat a0_re, a0_im;

   spinorFloat a1_re, a1_im;

   spinorFloat a2_re, a2_im;

   spinorFloat b0_re, b0_im;

   spinorFloat b1_re, b1_im;

   spinorFloat b2_re, b2_im;


 #ifdef MULTI_GPU

   if (kernel_type == INTERIOR_KERNEL) {

 #endif


     // read spinor from device memory

     READ_SPINOR(SPINORTEX, param.sp_stride, sp_idx, sp_idx);

 #ifdef TWIST_INV_DSLASH

     APPLY_TWIST_INV(-a, b, i);

 #endif


     // project spinor into half spinors

     a0_re = +i00_re-i30_re;

     a0_im = +i00_im-i30_im;

     a1_re = +i01_re-i31_re;

     a1_im = +i01_im-i31_im;

     a2_re = +i02_re-i32_re;

     a2_im = +i02_im-i32_im;

     b0_re = +i10_re+i20_re;

     b0_im = +i10_im+i20_im;

     b1_re = +i11_re+i21_re;

     b1_im = +i11_im+i21_im;

     b2_re = +i12_re+i22_re;

     b2_im = +i12_im+i22_im;


 #ifdef MULTI_GPU

   } else {


     const int sp_stride_pad = ghostFace[static_cast<int>(kernel_type)];


     // read half spinor from device memory

     READ_HALF_SPINOR(SPINORTEX, sp_stride_pad, sp_idx, sp_norm_idx);


     a0_re = i00_re;  a0_im = i00_im;

     a1_re = i01_re;  a1_im = i01_im;

     a2_re = i02_re;  a2_im = i02_im;

     b0_re = i10_re;  b0_im = i10_im;

     b1_re = i11_re;  b1_im = i11_im;

     b2_re = i12_re;  b2_im = i12_im;


   }

 #endif // MULTI_GPU


   // read gauge matrix from device memory

   READ_GAUGE_MATRIX(G, GAUGE1TEX, 3, ga_idx, ga_stride);


   // reconstruct gauge matrix

   RECONSTRUCT_GAUGE_MATRIX(3);


   // multiply row 0

   spinorFloat A0_re = 0;

   A0_re += gT00_re * a0_re;

   A0_re -= gT00_im * a0_im;

   A0_re += gT01_re * a1_re;

   A0_re -= gT01_im * a1_im;

   A0_re += gT02_re * a2_re;

   A0_re -= gT02_im * a2_im;

   spinorFloat A0_im = 0;

   A0_im += gT00_re * a0_im;

   A0_im += gT00_im * a0_re;

   A0_im += gT01_re * a1_im;

   A0_im += gT01_im * a1_re;

   A0_im += gT02_re * a2_im;

   A0_im += gT02_im * a2_re;

   spinorFloat B0_re = 0;

   B0_re += gT00_re * b0_re;

   B0_re -= gT00_im * b0_im;

   B0_re += gT01_re * b1_re;

   B0_re -= gT01_im * b1_im;

   B0_re += gT02_re * b2_re;

   B0_re -= gT02_im * b2_im;

   spinorFloat B0_im = 0;

   B0_im += gT00_re * b0_im;

   B0_im += gT00_im * b0_re;

   B0_im += gT01_re * b1_im;

   B0_im += gT01_im * b1_re;

   B0_im += gT02_re * b2_im;

   B0_im += gT02_im * b2_re;


   // multiply row 1

   spinorFloat A1_re = 0;

   A1_re += gT10_re * a0_re;

   A1_re -= gT10_im * a0_im;

   A1_re += gT11_re * a1_re;

   A1_re -= gT11_im * a1_im;

   A1_re += gT12_re * a2_re;

   A1_re -= gT12_im * a2_im;

   spinorFloat A1_im = 0;

   A1_im += gT10_re * a0_im;

   A1_im += gT10_im * a0_re;

   A1_im += gT11_re * a1_im;

   A1_im += gT11_im * a1_re;

   A1_im += gT12_re * a2_im;

   A1_im += gT12_im * a2_re;

   spinorFloat B1_re = 0;

   B1_re += gT10_re * b0_re;

   B1_re -= gT10_im * b0_im;

   B1_re += gT11_re * b1_re;

   B1_re -= gT11_im * b1_im;

   B1_re += gT12_re * b2_re;

   B1_re -= gT12_im * b2_im;

   spinorFloat B1_im = 0;

   B1_im += gT10_re * b0_im;

   B1_im += gT10_im * b0_re;

   B1_im += gT11_re * b1_im;

   B1_im += gT11_im * b1_re;

   B1_im += gT12_re * b2_im;

   B1_im += gT12_im * b2_re;


   // multiply row 2

   spinorFloat A2_re = 0;

   A2_re += gT20_re * a0_re;

   A2_re -= gT20_im * a0_im;

   A2_re += gT21_re * a1_re;

   A2_re -= gT21_im * a1_im;

   A2_re += gT22_re * a2_re;

   A2_re -= gT22_im * a2_im;

   spinorFloat A2_im = 0;

   A2_im += gT20_re * a0_im;

   A2_im += gT20_im * a0_re;

   A2_im += gT21_re * a1_im;

   A2_im += gT21_im * a1_re;

   A2_im += gT22_re * a2_im;

   A2_im += gT22_im * a2_re;

   spinorFloat B2_re = 0;

   B2_re += gT20_re * b0_re;

   B2_re -= gT20_im * b0_im;

   B2_re += gT21_re * b1_re;

   B2_re -= gT21_im * b1_im;

   B2_re += gT22_re * b2_re;

   B2_re -= gT22_im * b2_im;

   spinorFloat B2_im = 0;

   B2_im += gT20_re * b0_im;

   B2_im += gT20_im * b0_re;

   B2_im += gT21_re * b1_im;

   B2_im += gT21_im * b1_re;

   B2_im += gT22_re * b2_im;

   B2_im += gT22_im * b2_re;


   o00_re += A0_re;

   o00_im += A0_im;

   o10_re += B0_re;

   o10_im += B0_im;

   o20_re += B0_re;

   o20_im += B0_im;

   o30_re -= A0_re;

   o30_im -= A0_im;


   o01_re += A1_re;

   o01_im += A1_im;

   o11_re += B1_re;

   o11_im += B1_im;

   o21_re += B1_re;

   o21_im += B1_im;

   o31_re -= A1_re;

   o31_im -= A1_im;


   o02_re += A2_re;

   o02_im += A2_im;

   o12_re += B2_re;

   o12_im += B2_im;

   o22_re += B2_re;

   o22_im += B2_im;

   o32_re -= A2_re;

   o32_im -= A2_im;


 }


 #ifdef MULTI_GPU

 if ( (kernel_type == INTERIOR_KERNEL && (!param.ghostDim[2] || x3<X3m1)) ||

      (kernel_type == EXTERIOR_KERNEL_Z && x3==X3m1) )

 #endif

 {

   // Projector P2+

   // 1 0 i 0

   // 0 1 0 -i

   // -i 0 1 0

   // 0 i 0 1


 #ifdef MULTI_GPU

   const int sp_idx = (kernel_type == INTERIOR_KERNEL) ? (x3==X3m1 ? X-X3X2X1mX2X1 : X+X2X1) >> 1 :

     face_idx + param.ghostOffset[static_cast<int>(kernel_type)];

 #else

   const int sp_idx = (x3==X3m1 ? X-X3X2X1mX2X1 : X+X2X1) >> 1;

 #endif


   const int ga_idx = sid;


   spinorFloat a0_re, a0_im;

   spinorFloat a1_re, a1_im;

   spinorFloat a2_re, a2_im;

   spinorFloat b0_re, b0_im;

   spinorFloat b1_re, b1_im;

   spinorFloat b2_re, b2_im;


 #ifdef MULTI_GPU

   if (kernel_type == INTERIOR_KERNEL) {

 #endif


     // read spinor from device memory

     READ_SPINOR(SPINORTEX, param.sp_stride, sp_idx, sp_idx);

 #ifdef TWIST_INV_DSLASH

     APPLY_TWIST_INV(-a, b, i);

 #endif


     // project spinor into half spinors

     a0_re = +i00_re-i20_im;

     a0_im = +i00_im+i20_re;

     a1_re = +i01_re-i21_im;

     a1_im = +i01_im+i21_re;

     a2_re = +i02_re-i22_im;

     a2_im = +i02_im+i22_re;

     b0_re = +i10_re+i30_im;

     b0_im = +i10_im-i30_re;

     b1_re = +i11_re+i31_im;

     b1_im = +i11_im-i31_re;

     b2_re = +i12_re+i32_im;

     b2_im = +i12_im-i32_re;


 #ifdef MULTI_GPU

   } else {


     const int sp_stride_pad = ghostFace[static_cast<int>(kernel_type)];


     // read half spinor from device memory

     READ_HALF_SPINOR(SPINORTEX, sp_stride_pad, sp_idx + (SPINOR_HOP/2)*sp_stride_pad, sp_norm_idx);


     a0_re = i00_re;  a0_im = i00_im;

     a1_re = i01_re;  a1_im = i01_im;

     a2_re = i02_re;  a2_im = i02_im;

     b0_re = i10_re;  b0_im = i10_im;

     b1_re = i11_re;  b1_im = i11_im;

     b2_re = i12_re;  b2_im = i12_im;


   }

 #endif // MULTI_GPU


   // read gauge matrix from device memory

   READ_GAUGE_MATRIX(G, GAUGE0TEX, 4, ga_idx, ga_stride);


   // reconstruct gauge matrix

   RECONSTRUCT_GAUGE_MATRIX(4);


   // multiply row 0

   spinorFloat A0_re = 0;

   A0_re += g00_re * a0_re;

   A0_re -= g00_im * a0_im;

   A0_re += g01_re * a1_re;

   A0_re -= g01_im * a1_im;

   A0_re += g02_re * a2_re;

   A0_re -= g02_im * a2_im;

   spinorFloat A0_im = 0;

   A0_im += g00_re * a0_im;

   A0_im += g00_im * a0_re;

   A0_im += g01_re * a1_im;

   A0_im += g01_im * a1_re;

   A0_im += g02_re * a2_im;

   A0_im += g02_im * a2_re;

   spinorFloat B0_re = 0;

   B0_re += g00_re * b0_re;

   B0_re -= g00_im * b0_im;

   B0_re += g01_re * b1_re;

   B0_re -= g01_im * b1_im;

   B0_re += g02_re * b2_re;

   B0_re -= g02_im * b2_im;

   spinorFloat B0_im = 0;

   B0_im += g00_re * b0_im;

   B0_im += g00_im * b0_re;

   B0_im += g01_re * b1_im;

   B0_im += g01_im * b1_re;

   B0_im += g02_re * b2_im;

   B0_im += g02_im * b2_re;


   // multiply row 1

   spinorFloat A1_re = 0;

   A1_re += g10_re * a0_re;

   A1_re -= g10_im * a0_im;

   A1_re += g11_re * a1_re;

   A1_re -= g11_im * a1_im;

   A1_re += g12_re * a2_re;

   A1_re -= g12_im * a2_im;

   spinorFloat A1_im = 0;

   A1_im += g10_re * a0_im;

   A1_im += g10_im * a0_re;

   A1_im += g11_re * a1_im;

   A1_im += g11_im * a1_re;

   A1_im += g12_re * a2_im;

   A1_im += g12_im * a2_re;

   spinorFloat B1_re = 0;

   B1_re += g10_re * b0_re;

   B1_re -= g10_im * b0_im;

   B1_re += g11_re * b1_re;

   B1_re -= g11_im * b1_im;

   B1_re += g12_re * b2_re;

   B1_re -= g12_im * b2_im;

   spinorFloat B1_im = 0;

   B1_im += g10_re * b0_im;

   B1_im += g10_im * b0_re;

   B1_im += g11_re * b1_im;

   B1_im += g11_im * b1_re;

   B1_im += g12_re * b2_im;

   B1_im += g12_im * b2_re;


   // multiply row 2

   spinorFloat A2_re = 0;

   A2_re += g20_re * a0_re;

   A2_re -= g20_im * a0_im;

   A2_re += g21_re * a1_re;

   A2_re -= g21_im * a1_im;

   A2_re += g22_re * a2_re;

   A2_re -= g22_im * a2_im;

   spinorFloat A2_im = 0;

   A2_im += g20_re * a0_im;

   A2_im += g20_im * a0_re;

   A2_im += g21_re * a1_im;

   A2_im += g21_im * a1_re;

   A2_im += g22_re * a2_im;

   A2_im += g22_im * a2_re;

   spinorFloat B2_re = 0;

   B2_re += g20_re * b0_re;

   B2_re -= g20_im * b0_im;

   B2_re += g21_re * b1_re;

   B2_re -= g21_im * b1_im;

   B2_re += g22_re * b2_re;

   B2_re -= g22_im * b2_im;

   spinorFloat B2_im = 0;

   B2_im += g20_re * b0_im;

   B2_im += g20_im * b0_re;

   B2_im += g21_re * b1_im;

   B2_im += g21_im * b1_re;

   B2_im += g22_re * b2_im;

   B2_im += g22_im * b2_re;


   o00_re += A0_re;

   o00_im += A0_im;

   o10_re += B0_re;

   o10_im += B0_im;

   o20_re += A0_im;

   o20_im -= A0_re;

   o30_re -= B0_im;

   o30_im += B0_re;


   o01_re += A1_re;

   o01_im += A1_im;

   o11_re += B1_re;

   o11_im += B1_im;

   o21_re += A1_im;

   o21_im -= A1_re;

   o31_re -= B1_im;

   o31_im += B1_re;


   o02_re += A2_re;

   o02_im += A2_im;

   o12_re += B2_re;

   o12_im += B2_im;

   o22_re += A2_im;

   o22_im -= A2_re;

   o32_re -= B2_im;

   o32_im += B2_re;


 }


 #ifdef MULTI_GPU

 if ( (kernel_type == INTERIOR_KERNEL && (!param.ghostDim[2] || x3>0)) ||

      (kernel_type == EXTERIOR_KERNEL_Z && x3==0) )

 #endif

 {

   // Projector P2-

   // 1 0 -i 0

   // 0 1 0 i

   // i 0 1 0

   // 0 -i 0 1


 #ifdef MULTI_GPU

   const int sp_idx = (kernel_type == INTERIOR_KERNEL) ? (x3==0 ? X+X3X2X1mX2X1 : X-X2X1) >> 1 :

     face_idx + param.ghostOffset[static_cast<int>(kernel_type)];

 #else

   const int sp_idx = (x3==0 ? X+X3X2X1mX2X1 : X-X2X1) >> 1;

 #endif


 #ifdef MULTI_GPU

   const int ga_idx = ((kernel_type == INTERIOR_KERNEL) ? sp_idx : Vh+face_idx);

 #else

   const int ga_idx = sp_idx;

 #endif


   spinorFloat a0_re, a0_im;

   spinorFloat a1_re, a1_im;

   spinorFloat a2_re, a2_im;

   spinorFloat b0_re, b0_im;

   spinorFloat b1_re, b1_im;

   spinorFloat b2_re, b2_im;


 #ifdef MULTI_GPU

   if (kernel_type == INTERIOR_KERNEL) {

 #endif


     // read spinor from device memory

     READ_SPINOR(SPINORTEX, param.sp_stride, sp_idx, sp_idx);

 #ifdef TWIST_INV_DSLASH

     APPLY_TWIST_INV(-a, b, i);

 #endif


     // project spinor into half spinors

     a0_re = +i00_re+i20_im;

     a0_im = +i00_im-i20_re;

     a1_re = +i01_re+i21_im;

     a1_im = +i01_im-i21_re;

     a2_re = +i02_re+i22_im;

     a2_im = +i02_im-i22_re;

     b0_re = +i10_re-i30_im;

     b0_im = +i10_im+i30_re;

     b1_re = +i11_re-i31_im;

     b1_im = +i11_im+i31_re;

     b2_re = +i12_re-i32_im;

     b2_im = +i12_im+i32_re;


 #ifdef MULTI_GPU

   } else {


     const int sp_stride_pad = ghostFace[static_cast<int>(kernel_type)];


     // read half spinor from device memory

     READ_HALF_SPINOR(SPINORTEX, sp_stride_pad, sp_idx, sp_norm_idx);


     a0_re = i00_re;  a0_im = i00_im;

     a1_re = i01_re;  a1_im = i01_im;

     a2_re = i02_re;  a2_im = i02_im;

     b0_re = i10_re;  b0_im = i10_im;

     b1_re = i11_re;  b1_im = i11_im;

     b2_re = i12_re;  b2_im = i12_im;


   }

 #endif // MULTI_GPU


   // read gauge matrix from device memory

   READ_GAUGE_MATRIX(G, GAUGE1TEX, 5, ga_idx, ga_stride);


   // reconstruct gauge matrix

   RECONSTRUCT_GAUGE_MATRIX(5);


   // multiply row 0

   spinorFloat A0_re = 0;

   A0_re += gT00_re * a0_re;

   A0_re -= gT00_im * a0_im;

   A0_re += gT01_re * a1_re;

   A0_re -= gT01_im * a1_im;

   A0_re += gT02_re * a2_re;

   A0_re -= gT02_im * a2_im;

   spinorFloat A0_im = 0;

   A0_im += gT00_re * a0_im;

   A0_im += gT00_im * a0_re;

   A0_im += gT01_re * a1_im;

   A0_im += gT01_im * a1_re;

   A0_im += gT02_re * a2_im;

   A0_im += gT02_im * a2_re;

   spinorFloat B0_re = 0;

   B0_re += gT00_re * b0_re;

   B0_re -= gT00_im * b0_im;

   B0_re += gT01_re * b1_re;

   B0_re -= gT01_im * b1_im;

   B0_re += gT02_re * b2_re;

   B0_re -= gT02_im * b2_im;

   spinorFloat B0_im = 0;

   B0_im += gT00_re * b0_im;

   B0_im += gT00_im * b0_re;

   B0_im += gT01_re * b1_im;

   B0_im += gT01_im * b1_re;

   B0_im += gT02_re * b2_im;

   B0_im += gT02_im * b2_re;


   // multiply row 1

   spinorFloat A1_re = 0;

   A1_re += gT10_re * a0_re;

   A1_re -= gT10_im * a0_im;

   A1_re += gT11_re * a1_re;

   A1_re -= gT11_im * a1_im;

   A1_re += gT12_re * a2_re;

   A1_re -= gT12_im * a2_im;

   spinorFloat A1_im = 0;

   A1_im += gT10_re * a0_im;

   A1_im += gT10_im * a0_re;

   A1_im += gT11_re * a1_im;

   A1_im += gT11_im * a1_re;

   A1_im += gT12_re * a2_im;

   A1_im += gT12_im * a2_re;

   spinorFloat B1_re = 0;

   B1_re += gT10_re * b0_re;

   B1_re -= gT10_im * b0_im;

   B1_re += gT11_re * b1_re;

   B1_re -= gT11_im * b1_im;

   B1_re += gT12_re * b2_re;

   B1_re -= gT12_im * b2_im;

   spinorFloat B1_im = 0;

   B1_im += gT10_re * b0_im;

   B1_im += gT10_im * b0_re;

   B1_im += gT11_re * b1_im;

   B1_im += gT11_im * b1_re;

   B1_im += gT12_re * b2_im;

   B1_im += gT12_im * b2_re;


   // multiply row 2

   spinorFloat A2_re = 0;

   A2_re += gT20_re * a0_re;

   A2_re -= gT20_im * a0_im;

   A2_re += gT21_re * a1_re;

   A2_re -= gT21_im * a1_im;

   A2_re += gT22_re * a2_re;

   A2_re -= gT22_im * a2_im;

   spinorFloat A2_im = 0;

   A2_im += gT20_re * a0_im;

   A2_im += gT20_im * a0_re;

   A2_im += gT21_re * a1_im;

   A2_im += gT21_im * a1_re;

   A2_im += gT22_re * a2_im;

   A2_im += gT22_im * a2_re;

   spinorFloat B2_re = 0;

   B2_re += gT20_re * b0_re;

   B2_re -= gT20_im * b0_im;

   B2_re += gT21_re * b1_re;

   B2_re -= gT21_im * b1_im;

   B2_re += gT22_re * b2_re;

   B2_re -= gT22_im * b2_im;

   spinorFloat B2_im = 0;

   B2_im += gT20_re * b0_im;

   B2_im += gT20_im * b0_re;

   B2_im += gT21_re * b1_im;

   B2_im += gT21_im * b1_re;

   B2_im += gT22_re * b2_im;

   B2_im += gT22_im * b2_re;


   o00_re += A0_re;

   o00_im += A0_im;

   o10_re += B0_re;

   o10_im += B0_im;

   o20_re -= A0_im;

   o20_im += A0_re;

   o30_re += B0_im;

   o30_im -= B0_re;


   o01_re += A1_re;

   o01_im += A1_im;

   o11_re += B1_re;

   o11_im += B1_im;

   o21_re -= A1_im;

   o21_im += A1_re;

   o31_re += B1_im;

   o31_im -= B1_re;


   o02_re += A2_re;

   o02_im += A2_im;

   o12_re += B2_re;

   o12_im += B2_im;

   o22_re -= A2_im;

   o22_im += A2_re;

   o32_re += B2_im;

   o32_im -= B2_re;


 }


 #ifdef MULTI_GPU

 if ( (kernel_type == INTERIOR_KERNEL && (!param.ghostDim[3] || x4<X4m1)) ||

      (kernel_type == EXTERIOR_KERNEL_T && x4==X4m1) )

 #endif

 {

   // Projector P3+

   // 2 0 0 0

   // 0 2 0 0

   // 0 0 0 0

   // 0 0 0 0


 #ifdef MULTI_GPU

   const int sp_idx = (kernel_type == INTERIOR_KERNEL) ? (x4==X4m1 ? X-X4X3X2X1mX3X2X1 : X+X3X2X1) >> 1 :

     face_idx + param.ghostOffset[static_cast<int>(kernel_type)];

 #else

   const int sp_idx = (x4==X4m1 ? X-X4X3X2X1mX3X2X1 : X+X3X2X1) >> 1;

 #endif


   const int ga_idx = sid;


   if (gauge_fixed && ga_idx < X4X3X2X1hmX3X2X1h)

   {

     spinorFloat a0_re, a0_im;

     spinorFloat a1_re, a1_im;

     spinorFloat a2_re, a2_im;

     spinorFloat b0_re, b0_im;

     spinorFloat b1_re, b1_im;

     spinorFloat b2_re, b2_im;


 #ifdef MULTI_GPU

     if (kernel_type == INTERIOR_KERNEL) {

 #endif


       // read spinor from device memory

 #ifndef TWIST_INV_DSLASH

       READ_SPINOR_UP(SPINORTEX, param.sp_stride, sp_idx, sp_idx);

 #else

       READ_SPINOR(SPINORTEX, param.sp_stride, sp_idx, sp_idx);

       APPLY_TWIST_INV(-a, b, i);

 #endif


       // project spinor into half spinors

       a0_re = +2*i00_re;

       a0_im = +2*i00_im;

       a1_re = +2*i01_re;

       a1_im = +2*i01_im;

       a2_re = +2*i02_re;

       a2_im = +2*i02_im;

       b0_re = +2*i10_re;

       b0_im = +2*i10_im;

       b1_re = +2*i11_re;

       b1_im = +2*i11_im;

       b2_re = +2*i12_re;

       b2_im = +2*i12_im;


 #ifdef MULTI_GPU

     } else {


       const int sp_stride_pad = ghostFace[static_cast<int>(kernel_type)];

       //const int t_proj_scale = TPROJSCALE;


       // read half spinor from device memory

       READ_HALF_SPINOR(SPINORTEX, sp_stride_pad, sp_idx + (SPINOR_HOP/2)*sp_stride_pad, sp_norm_idx);


 #ifdef TWIST_INV_DSLASH

       a0_re = i00_re;  a0_im = i00_im;

       a1_re = i01_re;  a1_im = i01_im;

       a2_re = i02_re;  a2_im = i02_im;

       b0_re = i10_re;  b0_im = i10_im;

       b1_re = i11_re;  b1_im = i11_im;

       b2_re = i12_re;  b2_im = i12_im;

 #else

       a0_re = 2*i00_re;  a0_im = 2*i00_im;

       a1_re = 2*i01_re;  a1_im = 2*i01_im;

       a2_re = 2*i02_re;  a2_im = 2*i02_im;

       b0_re = 2*i10_re;  b0_im = 2*i10_im;

       b1_re = 2*i11_re;  b1_im = 2*i11_im;

       b2_re = 2*i12_re;  b2_im = 2*i12_im;

 #endif


     }

 #endif // MULTI_GPU


     // identity gauge matrix

     spinorFloat A0_re = a0_re; spinorFloat A0_im = a0_im;

     spinorFloat B0_re = b0_re; spinorFloat B0_im = b0_im;

     spinorFloat A1_re = a1_re; spinorFloat A1_im = a1_im;

     spinorFloat B1_re = b1_re; spinorFloat B1_im = b1_im;

     spinorFloat A2_re = a2_re; spinorFloat A2_im = a2_im;

     spinorFloat B2_re = b2_re; spinorFloat B2_im = b2_im;


     o00_re += A0_re;

     o00_im += A0_im;

     o10_re += B0_re;

     o10_im += B0_im;


     o01_re += A1_re;

     o01_im += A1_im;

     o11_re += B1_re;

     o11_im += B1_im;


     o02_re += A2_re;

     o02_im += A2_im;

     o12_re += B2_re;

     o12_im += B2_im;


   } else {

     spinorFloat a0_re, a0_im;

     spinorFloat a1_re, a1_im;

     spinorFloat a2_re, a2_im;

     spinorFloat b0_re, b0_im;

     spinorFloat b1_re, b1_im;

     spinorFloat b2_re, b2_im;


 #ifdef MULTI_GPU

     if (kernel_type == INTERIOR_KERNEL) {

 #endif


       // read spinor from device memory

 #ifndef TWIST_INV_DSLASH

       READ_SPINOR_UP(SPINORTEX, param.sp_stride, sp_idx, sp_idx);

 #else

       READ_SPINOR(SPINORTEX, param.sp_stride, sp_idx, sp_idx);

       APPLY_TWIST_INV(-a, b, i);

 #endif


       // project spinor into half spinors

       a0_re = +2*i00_re;

       a0_im = +2*i00_im;

       a1_re = +2*i01_re;

       a1_im = +2*i01_im;

       a2_re = +2*i02_re;

       a2_im = +2*i02_im;

       b0_re = +2*i10_re;

       b0_im = +2*i10_im;

       b1_re = +2*i11_re;

       b1_im = +2*i11_im;

       b2_re = +2*i12_re;

       b2_im = +2*i12_im;


 #ifdef MULTI_GPU

     } else {


       const int sp_stride_pad = ghostFace[static_cast<int>(kernel_type)];

       //const int t_proj_scale = TPROJSCALE;


       // read half spinor from device memory

       READ_HALF_SPINOR(SPINORTEX, sp_stride_pad, sp_idx + (SPINOR_HOP/2)*sp_stride_pad, sp_norm_idx);


 #ifdef TWIST_INV_DSLASH

       a0_re = i00_re;  a0_im = i00_im;

       a1_re = i01_re;  a1_im = i01_im;

       a2_re = i02_re;  a2_im = i02_im;

       b0_re = i10_re;  b0_im = i10_im;

       b1_re = i11_re;  b1_im = i11_im;

       b2_re = i12_re;  b2_im = i12_im;

 #else

       a0_re = 2*i00_re;  a0_im = 2*i00_im;

       a1_re = 2*i01_re;  a1_im = 2*i01_im;

       a2_re = 2*i02_re;  a2_im = 2*i02_im;

       b0_re = 2*i10_re;  b0_im = 2*i10_im;

       b1_re = 2*i11_re;  b1_im = 2*i11_im;

       b2_re = 2*i12_re;  b2_im = 2*i12_im;

 #endif


     }

 #endif // MULTI_GPU


     // read gauge matrix from device memory

     READ_GAUGE_MATRIX(G, GAUGE0TEX, 6, ga_idx, ga_stride);


     // reconstruct gauge matrix

     RECONSTRUCT_GAUGE_MATRIX(6);


     // multiply row 0

     spinorFloat A0_re = 0;

     A0_re += g00_re * a0_re;

     A0_re -= g00_im * a0_im;

     A0_re += g01_re * a1_re;

     A0_re -= g01_im * a1_im;

     A0_re += g02_re * a2_re;

     A0_re -= g02_im * a2_im;

     spinorFloat A0_im = 0;

     A0_im += g00_re * a0_im;

     A0_im += g00_im * a0_re;

     A0_im += g01_re * a1_im;

     A0_im += g01_im * a1_re;

     A0_im += g02_re * a2_im;

     A0_im += g02_im * a2_re;

     spinorFloat B0_re = 0;

     B0_re += g00_re * b0_re;

     B0_re -= g00_im * b0_im;

     B0_re += g01_re * b1_re;

     B0_re -= g01_im * b1_im;

     B0_re += g02_re * b2_re;

     B0_re -= g02_im * b2_im;

     spinorFloat B0_im = 0;

     B0_im += g00_re * b0_im;

     B0_im += g00_im * b0_re;

     B0_im += g01_re * b1_im;

     B0_im += g01_im * b1_re;

     B0_im += g02_re * b2_im;

     B0_im += g02_im * b2_re;


     // multiply row 1

     spinorFloat A1_re = 0;

     A1_re += g10_re * a0_re;

     A1_re -= g10_im * a0_im;

     A1_re += g11_re * a1_re;

     A1_re -= g11_im * a1_im;

     A1_re += g12_re * a2_re;

     A1_re -= g12_im * a2_im;

     spinorFloat A1_im = 0;

     A1_im += g10_re * a0_im;

     A1_im += g10_im * a0_re;

     A1_im += g11_re * a1_im;

     A1_im += g11_im * a1_re;

     A1_im += g12_re * a2_im;

     A1_im += g12_im * a2_re;

     spinorFloat B1_re = 0;

     B1_re += g10_re * b0_re;

     B1_re -= g10_im * b0_im;

     B1_re += g11_re * b1_re;

     B1_re -= g11_im * b1_im;

     B1_re += g12_re * b2_re;

     B1_re -= g12_im * b2_im;

     spinorFloat B1_im = 0;

     B1_im += g10_re * b0_im;

     B1_im += g10_im * b0_re;

     B1_im += g11_re * b1_im;

     B1_im += g11_im * b1_re;

     B1_im += g12_re * b2_im;

     B1_im += g12_im * b2_re;


     // multiply row 2

     spinorFloat A2_re = 0;

     A2_re += g20_re * a0_re;

     A2_re -= g20_im * a0_im;

     A2_re += g21_re * a1_re;

     A2_re -= g21_im * a1_im;

     A2_re += g22_re * a2_re;

     A2_re -= g22_im * a2_im;

     spinorFloat A2_im = 0;

     A2_im += g20_re * a0_im;

     A2_im += g20_im * a0_re;

     A2_im += g21_re * a1_im;

     A2_im += g21_im * a1_re;

     A2_im += g22_re * a2_im;

     A2_im += g22_im * a2_re;

     spinorFloat B2_re = 0;

     B2_re += g20_re * b0_re;

     B2_re -= g20_im * b0_im;

     B2_re += g21_re * b1_re;

     B2_re -= g21_im * b1_im;

     B2_re += g22_re * b2_re;

     B2_re -= g22_im * b2_im;

     spinorFloat B2_im = 0;

     B2_im += g20_re * b0_im;

     B2_im += g20_im * b0_re;

     B2_im += g21_re * b1_im;

     B2_im += g21_im * b1_re;

     B2_im += g22_re * b2_im;

     B2_im += g22_im * b2_re;


     o00_re += A0_re;

     o00_im += A0_im;

     o10_re += B0_re;

     o10_im += B0_im;


     o01_re += A1_re;

     o01_im += A1_im;

     o11_re += B1_re;

     o11_im += B1_im;


     o02_re += A2_re;

     o02_im += A2_im;

     o12_re += B2_re;

     o12_im += B2_im;


   }

 }


 #ifdef MULTI_GPU

 if ( (kernel_type == INTERIOR_KERNEL && (!param.ghostDim[3] || x4>0)) ||

      (kernel_type == EXTERIOR_KERNEL_T && x4==0) )

 #endif

 {

   // Projector P3-

   // 0 0 0 0

   // 0 0 0 0

   // 0 0 2 0

   // 0 0 0 2


 #ifdef MULTI_GPU

   const int sp_idx = (kernel_type == INTERIOR_KERNEL) ? (x4==0 ? X+X4X3X2X1mX3X2X1 : X-X3X2X1) >> 1 :

     face_idx + param.ghostOffset[static_cast<int>(kernel_type)];

 #else

   const int sp_idx = (x4==0 ? X+X4X3X2X1mX3X2X1 : X-X3X2X1) >> 1;

 #endif


 #ifdef MULTI_GPU

   const int ga_idx = ((kernel_type == INTERIOR_KERNEL) ? sp_idx : Vh+face_idx);

 #else

   const int ga_idx = sp_idx;

 #endif


   if (gauge_fixed && ga_idx < X4X3X2X1hmX3X2X1h)

   {

     spinorFloat a0_re, a0_im;

     spinorFloat a1_re, a1_im;

     spinorFloat a2_re, a2_im;

     spinorFloat b0_re, b0_im;

     spinorFloat b1_re, b1_im;

     spinorFloat b2_re, b2_im;


 #ifdef MULTI_GPU

     if (kernel_type == INTERIOR_KERNEL) {

 #endif


       // read spinor from device memory

 #ifndef TWIST_INV_DSLASH

       READ_SPINOR_DOWN(SPINORTEX, param.sp_stride, sp_idx, sp_idx);

 #else

       READ_SPINOR(SPINORTEX, param.sp_stride, sp_idx, sp_idx);

       APPLY_TWIST_INV(-a, b, i);

 #endif


       // project spinor into half spinors

       a0_re = +2*i20_re;

       a0_im = +2*i20_im;

       a1_re = +2*i21_re;

       a1_im = +2*i21_im;

       a2_re = +2*i22_re;

       a2_im = +2*i22_im;

       b0_re = +2*i30_re;

       b0_im = +2*i30_im;

       b1_re = +2*i31_re;

       b1_im = +2*i31_im;

       b2_re = +2*i32_re;

       b2_im = +2*i32_im;


 #ifdef MULTI_GPU

     } else {


       const int sp_stride_pad = ghostFace[static_cast<int>(kernel_type)];

       //const int t_proj_scale = TPROJSCALE;


       // read half spinor from device memory

       READ_HALF_SPINOR(SPINORTEX, sp_stride_pad, sp_idx, sp_norm_idx);


 #ifdef TWIST_INV_DSLASH

       a0_re = i00_re;  a0_im = i00_im;

       a1_re = i01_re;  a1_im = i01_im;

       a2_re = i02_re;  a2_im = i02_im;

       b0_re = i10_re;  b0_im = i10_im;

       b1_re = i11_re;  b1_im = i11_im;

       b2_re = i12_re;  b2_im = i12_im;

 #else

       a0_re = 2*i00_re;  a0_im = 2*i00_im;

       a1_re = 2*i01_re;  a1_im = 2*i01_im;

       a2_re = 2*i02_re;  a2_im = 2*i02_im;

       b0_re = 2*i10_re;  b0_im = 2*i10_im;

       b1_re = 2*i11_re;  b1_im = 2*i11_im;

       b2_re = 2*i12_re;  b2_im = 2*i12_im;

 #endif


     }

 #endif // MULTI_GPU


     // identity gauge matrix

     spinorFloat A0_re = a0_re; spinorFloat A0_im = a0_im;

     spinorFloat B0_re = b0_re; spinorFloat B0_im = b0_im;

     spinorFloat A1_re = a1_re; spinorFloat A1_im = a1_im;

     spinorFloat B1_re = b1_re; spinorFloat B1_im = b1_im;

     spinorFloat A2_re = a2_re; spinorFloat A2_im = a2_im;

     spinorFloat B2_re = b2_re; spinorFloat B2_im = b2_im;


     o20_re += A0_re;

     o20_im += A0_im;

     o30_re += B0_re;

     o30_im += B0_im;


     o21_re += A1_re;

     o21_im += A1_im;

     o31_re += B1_re;

     o31_im += B1_im;


     o22_re += A2_re;

     o22_im += A2_im;

     o32_re += B2_re;

     o32_im += B2_im;


   } else {

     spinorFloat a0_re, a0_im;

     spinorFloat a1_re, a1_im;

     spinorFloat a2_re, a2_im;

     spinorFloat b0_re, b0_im;

     spinorFloat b1_re, b1_im;

     spinorFloat b2_re, b2_im;


 #ifdef MULTI_GPU

     if (kernel_type == INTERIOR_KERNEL) {

 #endif


       // read spinor from device memory

 #ifndef TWIST_INV_DSLASH

       READ_SPINOR_DOWN(SPINORTEX, param.sp_stride, sp_idx, sp_idx);

 #else

       READ_SPINOR(SPINORTEX, param.sp_stride, sp_idx, sp_idx);

       APPLY_TWIST_INV(-a, b, i);

 #endif


       // project spinor into half spinors

       a0_re = +2*i20_re;

       a0_im = +2*i20_im;

       a1_re = +2*i21_re;

       a1_im = +2*i21_im;

       a2_re = +2*i22_re;

       a2_im = +2*i22_im;

       b0_re = +2*i30_re;

       b0_im = +2*i30_im;

       b1_re = +2*i31_re;

       b1_im = +2*i31_im;

       b2_re = +2*i32_re;

       b2_im = +2*i32_im;


 #ifdef MULTI_GPU

     } else {


       const int sp_stride_pad = ghostFace[static_cast<int>(kernel_type)];

       //const int t_proj_scale = TPROJSCALE;


       // read half spinor from device memory

       READ_HALF_SPINOR(SPINORTEX, sp_stride_pad, sp_idx, sp_norm_idx);


 #ifdef TWIST_INV_DSLASH

       a0_re = i00_re;  a0_im = i00_im;

       a1_re = i01_re;  a1_im = i01_im;

       a2_re = i02_re;  a2_im = i02_im;

       b0_re = i10_re;  b0_im = i10_im;

       b1_re = i11_re;  b1_im = i11_im;

       b2_re = i12_re;  b2_im = i12_im;

 #else

       a0_re = 2*i00_re;  a0_im = 2*i00_im;

       a1_re = 2*i01_re;  a1_im = 2*i01_im;

       a2_re = 2*i02_re;  a2_im = 2*i02_im;

       b0_re = 2*i10_re;  b0_im = 2*i10_im;

       b1_re = 2*i11_re;  b1_im = 2*i11_im;

       b2_re = 2*i12_re;  b2_im = 2*i12_im;

 #endif


     }

 #endif // MULTI_GPU


     // read gauge matrix from device memory

     READ_GAUGE_MATRIX(G, GAUGE1TEX, 7, ga_idx, ga_stride);


     // reconstruct gauge matrix

     RECONSTRUCT_GAUGE_MATRIX(7);


     // multiply row 0

     spinorFloat A0_re = 0;

     A0_re += gT00_re * a0_re;

     A0_re -= gT00_im * a0_im;

     A0_re += gT01_re * a1_re;

     A0_re -= gT01_im * a1_im;

     A0_re += gT02_re * a2_re;

     A0_re -= gT02_im * a2_im;

     spinorFloat A0_im = 0;

     A0_im += gT00_re * a0_im;

     A0_im += gT00_im * a0_re;

     A0_im += gT01_re * a1_im;

     A0_im += gT01_im * a1_re;

     A0_im += gT02_re * a2_im;

     A0_im += gT02_im * a2_re;

     spinorFloat B0_re = 0;

     B0_re += gT00_re * b0_re;

     B0_re -= gT00_im * b0_im;

     B0_re += gT01_re * b1_re;

     B0_re -= gT01_im * b1_im;

     B0_re += gT02_re * b2_re;

     B0_re -= gT02_im * b2_im;

     spinorFloat B0_im = 0;

     B0_im += gT00_re * b0_im;

     B0_im += gT00_im * b0_re;

     B0_im += gT01_re * b1_im;

     B0_im += gT01_im * b1_re;

     B0_im += gT02_re * b2_im;

     B0_im += gT02_im * b2_re;


     // multiply row 1

     spinorFloat A1_re = 0;

     A1_re += gT10_re * a0_re;

     A1_re -= gT10_im * a0_im;

     A1_re += gT11_re * a1_re;

     A1_re -= gT11_im * a1_im;

     A1_re += gT12_re * a2_re;

     A1_re -= gT12_im * a2_im;

     spinorFloat A1_im = 0;

     A1_im += gT10_re * a0_im;

     A1_im += gT10_im * a0_re;

     A1_im += gT11_re * a1_im;

     A1_im += gT11_im * a1_re;

     A1_im += gT12_re * a2_im;

     A1_im += gT12_im * a2_re;

     spinorFloat B1_re = 0;

     B1_re += gT10_re * b0_re;

     B1_re -= gT10_im * b0_im;

     B1_re += gT11_re * b1_re;

     B1_re -= gT11_im * b1_im;

     B1_re += gT12_re * b2_re;

     B1_re -= gT12_im * b2_im;

     spinorFloat B1_im = 0;

     B1_im += gT10_re * b0_im;

     B1_im += gT10_im * b0_re;

     B1_im += gT11_re * b1_im;

     B1_im += gT11_im * b1_re;

     B1_im += gT12_re * b2_im;

     B1_im += gT12_im * b2_re;


     // multiply row 2

     spinorFloat A2_re = 0;

     A2_re += gT20_re * a0_re;

     A2_re -= gT20_im * a0_im;

     A2_re += gT21_re * a1_re;

     A2_re -= gT21_im * a1_im;

     A2_re += gT22_re * a2_re;

     A2_re -= gT22_im * a2_im;

     spinorFloat A2_im = 0;

     A2_im += gT20_re * a0_im;

     A2_im += gT20_im * a0_re;

     A2_im += gT21_re * a1_im;

     A2_im += gT21_im * a1_re;

     A2_im += gT22_re * a2_im;

     A2_im += gT22_im * a2_re;

     spinorFloat B2_re = 0;

     B2_re += gT20_re * b0_re;

     B2_re -= gT20_im * b0_im;

     B2_re += gT21_re * b1_re;

     B2_re -= gT21_im * b1_im;

     B2_re += gT22_re * b2_re;

     B2_re -= gT22_im * b2_im;

     spinorFloat B2_im = 0;

     B2_im += gT20_re * b0_im;

     B2_im += gT20_im * b0_re;

     B2_im += gT21_re * b1_im;

     B2_im += gT21_im * b1_re;

     B2_im += gT22_re * b2_im;

     B2_im += gT22_im * b2_re;


     o20_re += A0_re;

     o20_im += A0_im;

     o30_re += B0_re;

     o30_im += B0_im;


     o21_re += A1_re;

     o21_im += A1_im;

     o31_re += B1_re;

     o31_im += B1_im;


     o22_re += A2_re;

     o22_im += A2_im;

     o32_re += B2_re;

     o32_im += B2_im;


   }

 }


 #ifdef MULTI_GPU


 int incomplete = 0; // Have all 8 contributions been computed for this site?


 switch(kernel_type) { // intentional fall-through

 case INTERIOR_KERNEL:

   incomplete = incomplete || (param.commDim[3] && (x4==0 || x4==X4m1));

 case EXTERIOR_KERNEL_T:

   incomplete = incomplete || (param.commDim[2] && (x3==0 || x3==X3m1));

 case EXTERIOR_KERNEL_Z:

   incomplete = incomplete || (param.commDim[1] && (x2==0 || x2==X2m1));

 case EXTERIOR_KERNEL_Y:

   incomplete = incomplete || (param.commDim[0] && (x1==0 || x1==X1m1));

 }


 if (!incomplete)

 #endif // MULTI_GPU

 {

 #ifdef DSLASH_XPAY

   READ_ACCUM(ACCUMTEX, param.sp_stride)


 #ifndef TWIST_XPAY

 #ifndef TWIST_INV_DSLASH

   //perform invert twist first:

   APPLY_TWIST_INV(-a, b, o);

 #endif

   o00_re += acc00_re;

   o00_im += acc00_im;

   o01_re += acc01_re;

   o01_im += acc01_im;

   o02_re += acc02_re;

   o02_im += acc02_im;

   o10_re += acc10_re;

   o10_im += acc10_im;

   o11_re += acc11_re;

   o11_im += acc11_im;

   o12_re += acc12_re;

   o12_im += acc12_im;

   o20_re += acc20_re;

   o20_im += acc20_im;

   o21_re += acc21_re;

   o21_im += acc21_im;

   o22_re += acc22_re;

   o22_im += acc22_im;

   o30_re += acc30_re;

   o30_im += acc30_im;

   o31_re += acc31_re;

   o31_im += acc31_im;

   o32_re += acc32_re;

   o32_im += acc32_im;

 #else

   APPLY_TWIST(-a, acc);

   //warning! b is unrelated to the twisted mass parameter in this case!


   o00_re = b*o00_re+acc00_re;

   o00_im = b*o00_im+acc00_im;

   o01_re = b*o01_re+acc01_re;

   o01_im = b*o01_im+acc01_im;

   o02_re = b*o02_re+acc02_re;

   o02_im = b*o02_im+acc02_im;

   o10_re = b*o10_re+acc10_re;

   o10_im = b*o10_im+acc10_im;

   o11_re = b*o11_re+acc11_re;

   o11_im = b*o11_im+acc11_im;

   o12_re = b*o12_re+acc12_re;

   o12_im = b*o12_im+acc12_im;

   o20_re = b*o20_re+acc20_re;

   o20_im = b*o20_im+acc20_im;

   o21_re = b*o21_re+acc21_re;

   o21_im = b*o21_im+acc21_im;

   o22_re = b*o22_re+acc22_re;

   o22_im = b*o22_im+acc22_im;

   o30_re = b*o30_re+acc30_re;

   o30_im = b*o30_im+acc30_im;

   o31_re = b*o31_re+acc31_re;

   o31_im = b*o31_im+acc31_im;

   o32_re = b*o32_re+acc32_re;

   o32_im = b*o32_im+acc32_im;

 #endif//TWIST_XPAY

 #else //no XPAY

 #ifndef TWIST_INV_DSLASH

      APPLY_TWIST_INV(-a, b, o);

 #endif

 #endif

 }


 // write spinor field back to device memory

 WRITE_SPINOR(param.sp_stride);


 // undefine to prevent warning when precision is changed

 #undef spinorFloat

 #undef SHARED_STRIDE


 #undef g00_re

 #undef g00_im

 #undef g01_re

 #undef g01_im

 #undef g02_re

 #undef g02_im

 #undef g10_re

 #undef g10_im

 #undef g11_re

 #undef g11_im

 #undef g12_re

 #undef g12_im

 #undef g20_re

 #undef g20_im

 #undef g21_re

 #undef g21_im

 #undef g22_re

 #undef g22_im


 #undef i00_re

 #undef i00_im

 #undef i01_re

 #undef i01_im

 #undef i02_re

 #undef i02_im

 #undef i10_re

 #undef i10_im

 #undef i11_re

 #undef i11_im

 #undef i12_re

 #undef i12_im

 #undef i20_re

 #undef i20_im

 #undef i21_re

 #undef i21_im

 #undef i22_re

 #undef i22_im

 #undef i30_re

 #undef i30_im

 #undef i31_re

 #undef i31_im

 #undef i32_re

 #undef i32_im


 #undef acc00_re

 #undef acc00_im

 #undef acc01_re

 #undef acc01_im

 #undef acc02_re

 #undef acc02_im

 #undef acc10_re

 #undef acc10_im

 #undef acc11_re

 #undef acc11_im

 #undef acc12_re

 #undef acc12_im

 #undef acc20_re

 #undef acc20_im

 #undef acc21_re

 #undef acc21_im

 #undef acc22_re

 #undef acc22_im

 #undef acc30_re

 #undef acc30_im

 #undef acc31_re

 #undef acc31_im

 #undef acc32_re

 #undef acc32_im


 #undef o00_re

 #undef o00_im

 #undef o01_re

 #undef o01_im

 #undef o02_re

 #undef o02_im

 #undef o10_re

 #undef o10_im

 #undef o11_re

 #undef o11_im

 #undef o12_re

 #undef o12_im

 #undef o20_re

 #undef o20_im

 #undef o21_re

 #undef o21_im

 #undef o22_re

 #undef o22_im

 #undef o30_re


 #undef VOLATILE

B1_re
spinorFloat B1_re
Definition: tm_dslash_dagger_g80_core.h:411

gT01_im
#define gT01_im
Definition: tm_dslash_dagger_g80_core.h:161

x1
int x1
Definition: tm_dslash_dagger_g80_core.h:217

gT21_re
#define gT21_re
Definition: tm_dslash_dagger_g80_core.h:172

Vh
__constant__ int Vh
Definition: dslash_constants.h:152

WRITE_SPINOR
WRITE_SPINOR(param.sp_stride)

acc12_re
#define acc12_re
Definition: tm_dslash_dagger_g80_core.h:98

gT10_re
#define gT10_re
Definition: tm_dslash_dagger_g80_core.h:164

acc22_re
#define acc22_re
Definition: tm_dslash_dagger_g80_core.h:104

x4
int x4
Definition: tm_dslash_dagger_g80_core.h:217

RECONSTRUCT_GAUGE_MATRIX
RECONSTRUCT_GAUGE_MATRIX(0)

X2
__constant__ int X2
Definition: dslash_constants.h:107

s
VOLATILE spinorFloat * s
Definition: tm_dslash_dagger_g80_core.h:211

APPLY_TWIST
#define APPLY_TWIST(a, reg)
Definition: io_spinor.h:1187

i32_re
#define i32_re
Definition: tm_dslash_dagger_g80_core.h:86

i11_re
#define i11_re
Definition: tm_dslash_dagger_g80_core.h:72

gT12_im
#define gT12_im
Definition: tm_dslash_dagger_g80_core.h:169

X2X1mX1
__constant__ int X2X1mX1
Definition: dslash_constants.h:126

acc02_im
#define acc02_im
Definition: tm_dslash_dagger_g80_core.h:93

EXTERIOR_KERNEL_X
Definition: dslash_constants.h:4

gT20_im
#define gT20_im
Definition: tm_dslash_dagger_g80_core.h:171

o32_re
VOLATILE spinorFloat o32_re
Definition: tm_dslash_dagger_g80_core.h:200

g00_im
#define g00_im
Definition: tm_dslash_dagger_g80_core.h:137

B0_im
spinorFloat B0_im
Definition: tm_dslash_dagger_g80_core.h:388

acc01_re
#define acc01_re
Definition: tm_dslash_dagger_g80_core.h:90

APPLY_TWIST_INV
#define APPLY_TWIST_INV(a, b, reg)
**************************only for deg tm:*******************************
Definition: io_spinor.h:1122

X3X2X1mX2X1
__constant__ int X3X2X1mX2X1
Definition: dslash_constants.h:127

A1_re
spinorFloat A1_re
Definition: tm_dslash_dagger_g80_core.h:397

acc30_re
#define acc30_re
Definition: tm_dslash_dagger_g80_core.h:106

o22_re
#define o22_re
Definition: tm_dslash_dagger_g80_core.h:194

X1
__constant__ int X1
Definition: dslash_constants.h:106

A1_im
spinorFloat A1_im
Definition: tm_dslash_dagger_g80_core.h:404

READ_INTERMEDIATE_SPINOR
#define READ_INTERMEDIATE_SPINOR
Definition: covDev.h:144

B2_re
spinorFloat B2_re
Definition: tm_dslash_dagger_g80_core.h:441

sp_idx
int sp_idx
Definition: dw_dslash5inv_core.h:157

g21_im
#define g21_im
Definition: tm_dslash_dagger_g80_core.h:151

gT11_im
#define gT11_im
Definition: tm_dslash_dagger_g80_core.h:167

READ_SPINOR
READ_SPINOR(SPINORTEX, param.sp_stride, sp_idx, sp_idx)

a2_im
spinorFloat a2_im
Definition: tm_dslash_dagger_g80_core.h:313

READ_SPINOR_DOWN
READ_SPINOR_DOWN(SPINORTEX, param.sp_stride, sp_idx, sp_idx)

face_idx
int face_idx
Definition: tm_ndeg_fused_exterior_dslash_core.h:196

x3
int x3
Definition: tm_dslash_dagger_g80_core.h:217

o11_re
#define o11_re
Definition: tm_dslash_dagger_g80_core.h:186

o20_re
#define o20_re
Definition: tm_dslash_dagger_g80_core.h:190

acc20_re
#define acc20_re
Definition: tm_dslash_dagger_g80_core.h:100

A2_im
spinorFloat A2_im
Definition: tm_dslash_dagger_g80_core.h:434

gT20_re
#define gT20_re
Definition: tm_dslash_dagger_g80_core.h:170

i21_im
#define i21_im
Definition: tm_dslash_dagger_g80_core.h:79

A0_im
spinorFloat A0_im
Definition: tm_dslash_dagger_g80_core.h:374

X3X2X1
__constant__ int X3X2X1
Definition: dslash_constants.h:139

i12_im
#define i12_im
Definition: tm_dslash_dagger_g80_core.h:75

gT02_im
#define gT02_im
Definition: tm_dslash_dagger_g80_core.h:163

B2_im
spinorFloat B2_im
Definition: tm_dslash_dagger_g80_core.h:448

i22_im
#define i22_im
Definition: tm_dslash_dagger_g80_core.h:81

dims
const int dims[]
Definition: tm_dslash_dagger_g80_core.h:235

i02_im
#define i02_im
Definition: tm_dslash_dagger_g80_core.h:69

READ_GAUGE_MATRIX
READ_GAUGE_MATRIX(G, GAUGE0TEX, 0, ga_idx, ga_stride)

g00_re
#define g00_re
Definition: tm_dslash_dagger_g80_core.h:136

a0_im
spinorFloat a0_im
Definition: tm_dslash_dagger_g80_core.h:311

gT10_im
#define gT10_im
Definition: tm_dslash_dagger_g80_core.h:165

i10_im
#define i10_im
Definition: tm_dslash_dagger_g80_core.h:71

acc01_im
#define acc01_im
Definition: tm_dslash_dagger_g80_core.h:91

acc31_im
#define acc31_im
Definition: tm_dslash_dagger_g80_core.h:109

g01_re
#define g01_re
Definition: tm_dslash_dagger_g80_core.h:138

i20_im
#define i20_im
Definition: tm_dslash_dagger_g80_core.h:77

gT21_im
#define gT21_im
Definition: tm_dslash_dagger_g80_core.h:173

b1_re
spinorFloat b1_re
Definition: tm_dslash_dagger_g80_core.h:315

i30_re
#define i30_re
Definition: tm_dslash_dagger_g80_core.h:82

g01_im
#define g01_im
Definition: tm_dslash_dagger_g80_core.h:139

i01_im
#define i01_im
Definition: tm_dslash_dagger_g80_core.h:67

a1_im
spinorFloat a1_im
Definition: tm_dslash_dagger_g80_core.h:312

acc10_re
#define acc10_re
Definition: tm_dslash_dagger_g80_core.h:94

param
QudaGaugeParam param
Definition: pack_test.cpp:17

acc00_im
#define acc00_im
Definition: tm_dslash_dagger_g80_core.h:89

o32_im
VOLATILE spinorFloat o32_im
Definition: tm_dslash_dagger_g80_core.h:201

ghostFace
__constant__ int ghostFace[QUDA_MAX_DIM+1]
Definition: dslash_constants.h:158

b1_im
spinorFloat b1_im
Definition: tm_dslash_dagger_g80_core.h:315

gT02_re
#define gT02_re
Definition: tm_dslash_dagger_g80_core.h:162

b0_re
spinorFloat b0_re
Definition: tm_dslash_dagger_g80_core.h:314

gT01_re
#define gT01_re
Definition: tm_dslash_dagger_g80_core.h:160

b0_im
spinorFloat b0_im
Definition: tm_dslash_dagger_g80_core.h:314

g10_im
#define g10_im
Definition: tm_dslash_dagger_g80_core.h:143

o00_re
#define o00_re
Definition: tm_dslash_dagger_g80_core.h:178

b2_im
spinorFloat b2_im
Definition: tm_dslash_dagger_g80_core.h:316

i10_re
#define i10_re
Definition: tm_dslash_dagger_g80_core.h:70

a0_re
spinorFloat a0_re
Definition: tm_dslash_dagger_g80_core.h:311

EXTERIOR_KERNEL_Y
Definition: dslash_constants.h:5

acc22_im
#define acc22_im
Definition: tm_dslash_dagger_g80_core.h:105

DSLASH_SHARED_FLOATS_PER_THREAD
#define DSLASH_SHARED_FLOATS_PER_THREAD
Definition: tm_dslash_dagger_g80_core.h:3

o11_im
#define o11_im
Definition: tm_dslash_dagger_g80_core.h:187

i22_re
#define i22_re
Definition: tm_dslash_dagger_g80_core.h:80

i00_im
#define i00_im
Definition: tm_dslash_dagger_g80_core.h:65

GAUGE0TEX
#define GAUGE0TEX
Definition: covDev.h:112

sid
int sid
Definition: tm_dslash_dagger_g80_core.h:224

g20_re
#define g20_re
Definition: tm_dslash_dagger_g80_core.h:148

acc11_re
#define acc11_re
Definition: tm_dslash_dagger_g80_core.h:96

i31_re
#define i31_re
Definition: tm_dslash_dagger_g80_core.h:84

dim
int dim
Definition: tm_ndeg_fused_exterior_dslash_core.h:195

B1_im
spinorFloat B1_im
Definition: tm_dslash_dagger_g80_core.h:418

gT00_im
#define gT00_im
Definition: tm_dslash_dagger_g80_core.h:159

o21_im
#define o21_im
Definition: tm_dslash_dagger_g80_core.h:193

i32_im
#define i32_im
Definition: tm_dslash_dagger_g80_core.h:87

acc11_im
#define acc11_im
Definition: tm_dslash_dagger_g80_core.h:97

g12_re
#define g12_re
Definition: tm_dslash_dagger_g80_core.h:146

acc21_re
#define acc21_re
Definition: tm_dslash_dagger_g80_core.h:102

io_spinor.h

X2m1
__constant__ int X2m1
Definition: dslash_constants.h:117

o02_re
#define o02_re
Definition: tm_dslash_dagger_g80_core.h:182

gT11_re
#define gT11_re
Definition: tm_dslash_dagger_g80_core.h:166

acc12_im
#define acc12_im
Definition: tm_dslash_dagger_g80_core.h:99

INTERIOR_KERNEL
Definition: dslash_constants.h:2

acc21_im
#define acc21_im
Definition: tm_dslash_dagger_g80_core.h:103

SPINORTEX
#define SPINORTEX
Definition: clover_def.h:40

o10_im
#define o10_im
Definition: tm_dslash_dagger_g80_core.h:185

gauge_fixed
__constant__ int gauge_fixed
Definition: dslash_constants.h:164

i01_re
#define i01_re
Definition: tm_dslash_dagger_g80_core.h:66

g22_im
#define g22_im
Definition: tm_dslash_dagger_g80_core.h:153

g22_re
#define g22_re
Definition: tm_dslash_dagger_g80_core.h:152

acc02_re
#define acc02_re
Definition: tm_dslash_dagger_g80_core.h:92

EXTERIOR_KERNEL_Z
Definition: dslash_constants.h:6

X4X3X2X1mX3X2X1
__constant__ int X4X3X2X1mX3X2X1
Definition: dslash_constants.h:128

g12_im
#define g12_im
Definition: tm_dslash_dagger_g80_core.h:147

s_data
__shared__ char s_data[]

SPINOR_HOP
#define SPINOR_HOP
Definition: covDev.h:158

o01_re
#define o01_re
Definition: tm_dslash_dagger_g80_core.h:180

acc32_re
#define acc32_re
Definition: tm_dslash_dagger_g80_core.h:110

A2_re
spinorFloat A2_re
Definition: tm_dslash_dagger_g80_core.h:427

spinorFloat
#define spinorFloat
Definition: tm_dslash_dagger_g80_core.h:63

READ_SPINOR_UP
READ_SPINOR_UP(SPINORTEX, param.sp_stride, sp_idx, sp_idx)

gT12_re
#define gT12_re
Definition: tm_dslash_dagger_g80_core.h:168

ga_stride
__constant__ int ga_stride
Definition: dslash_constants.h:156

g10_re
#define g10_re
Definition: tm_dslash_dagger_g80_core.h:142

b2_re
spinorFloat b2_re
Definition: tm_dslash_dagger_g80_core.h:316

o02_im
#define o02_im
Definition: tm_dslash_dagger_g80_core.h:183

acc31_re
#define acc31_re
Definition: tm_dslash_dagger_g80_core.h:108

X1m1
__constant__ int X1m1
Definition: dslash_constants.h:116

EXTERIOR_KERNEL_T
Definition: dslash_constants.h:7

o31_im
VOLATILE spinorFloat o31_im
Definition: tm_dslash_dagger_g80_core.h:199

acc20_im
#define acc20_im
Definition: tm_dslash_dagger_g80_core.h:101

gT22_im
#define gT22_im
Definition: tm_dslash_dagger_g80_core.h:175

X3
__constant__ int X3
Definition: dslash_constants.h:108

o22_im
#define o22_im
Definition: tm_dslash_dagger_g80_core.h:195

o12_re
#define o12_re
Definition: tm_dslash_dagger_g80_core.h:188

i12_re
#define i12_re
Definition: tm_dslash_dagger_g80_core.h:74

x2
int x2
Definition: tm_dslash_dagger_g80_core.h:217

o12_im
#define o12_im
Definition: tm_dslash_dagger_g80_core.h:189

i21_re
#define i21_re
Definition: tm_dslash_dagger_g80_core.h:78

i31_im
#define i31_im
Definition: tm_dslash_dagger_g80_core.h:85

acc30_im
#define acc30_im
Definition: tm_dslash_dagger_g80_core.h:107

ga_idx
const int ga_idx
Definition: tm_dslash_dagger_g80_core.h:309

A0_re
spinorFloat A0_re
Definition: tm_dslash_dagger_g80_core.h:367

i20_re
#define i20_re
Definition: tm_dslash_dagger_g80_core.h:76

o01_im
#define o01_im
Definition: tm_dslash_dagger_g80_core.h:181

X
int X
Definition: tm_dslash_dagger_g80_core.h:218

GAUGE1TEX
#define GAUGE1TEX
Definition: covDev.h:113

o30_im
VOLATILE spinorFloat o30_im
Definition: tm_dslash_dagger_g80_core.h:197

g02_im
#define g02_im
Definition: tm_dslash_dagger_g80_core.h:141

acc00_re
#define acc00_re
Definition: tm_dslash_dagger_g80_core.h:88

g02_re
#define g02_re
Definition: tm_dslash_dagger_g80_core.h:140

X4m1
__constant__ int X4m1
Definition: dslash_constants.h:119

gT22_re
#define gT22_re
Definition: tm_dslash_dagger_g80_core.h:174

coordsFromIndex< EVEN_X >
coordsFromIndex< EVEN_X >(X, x1, x2, x3, x4, sid, param.parity, dims)

VOLATILE
#define VOLATILE
Definition: tm_dslash_dagger_g80_core.h:9

gT00_re
#define gT00_re
Definition: tm_dslash_dagger_g80_core.h:158

o00_im
#define o00_im
Definition: tm_dslash_dagger_g80_core.h:179

g20_im
#define g20_im
Definition: tm_dslash_dagger_g80_core.h:149

i11_im
#define i11_im
Definition: tm_dslash_dagger_g80_core.h:73

read_gauge.h

g11_re
#define g11_re
Definition: tm_dslash_dagger_g80_core.h:144

g11_im
#define g11_im
Definition: tm_dslash_dagger_g80_core.h:145

READ_HALF_SPINOR
#define READ_HALF_SPINOR
Definition: io_spinor.h:390

INTERTEX
#define INTERTEX
Definition: covDev.h:149

o30_re
#define o30_re
Definition: tm_dslash_dagger_g80_core.h:196

B0_re
spinorFloat B0_re
Definition: tm_dslash_dagger_g80_core.h:381

X4X3X2X1hmX3X2X1h
__constant__ int X4X3X2X1hmX3X2X1h
Definition: dslash_constants.h:129

o21_re
#define o21_re
Definition: tm_dslash_dagger_g80_core.h:192

a2_re
spinorFloat a2_re
Definition: tm_dslash_dagger_g80_core.h:313

a1_re
spinorFloat a1_re
Definition: tm_dslash_dagger_g80_core.h:312

o10_re
#define o10_re
Definition: tm_dslash_dagger_g80_core.h:184

g21_re
#define g21_re
Definition: tm_dslash_dagger_g80_core.h:150

o31_re
VOLATILE spinorFloat o31_re
Definition: tm_dslash_dagger_g80_core.h:198

kernel_type
KernelType kernel_type
Definition: tm_ndeg_fused_exterior_dslash_core.h:13

i00_re
#define i00_re
Definition: tm_dslash_dagger_g80_core.h:64

o20_im
#define o20_im
Definition: tm_dslash_dagger_g80_core.h:191

X4
__constant__ int X4
Definition: dslash_constants.h:109

i02_re
#define i02_re
Definition: tm_dslash_dagger_g80_core.h:68

X3m1
__constant__ int X3m1
Definition: dslash_constants.h:118

i30_im
#define i30_im
Definition: tm_dslash_dagger_g80_core.h:83

SHARED_STRIDE
#define SHARED_STRIDE
Definition: tm_dslash_dagger_g80_core.h:206

X2X1
__constant__ int X2X1
Definition: dslash_constants.h:136

acc32_im
#define acc32_im
Definition: tm_dslash_dagger_g80_core.h:111

acc10_im
#define acc10_im
Definition: tm_dslash_dagger_g80_core.h:95