MatrixProductMMA.h
Go to the documentation of this file.
1 // This file is part of Eigen, a lightweight C++ template library
2 // for linear algebra.
3 //
4 // Copyright (C) 2020 Everton Constantino (everton.constantino@ibm.com)
5 // Copyright (C) 2021 Chip Kerchner (chip.kerchner@ibm.com)
6 //
7 // This Source Code Form is subject to the terms of the Mozilla
8 // Public License v. 2.0. If a copy of the MPL was not distributed
9 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
10 
11 #ifndef EIGEN_MATRIX_PRODUCT_MMA_ALTIVEC_H
12 #define EIGEN_MATRIX_PRODUCT_MMA_ALTIVEC_H
13 
14 #pragma GCC target("cpu=power10")
15 
16 #ifdef __has_builtin
17 #if !__has_builtin(__builtin_vsx_assemble_pair)
18 #define __builtin_vsx_assemble_pair __builtin_mma_assemble_pair
19 #endif
20 #endif
21 
22 namespace Eigen {
23 
24 namespace internal {
25 
26 template<typename Scalar, typename Packet>
27 EIGEN_ALWAYS_INLINE void bsetzeroMMA(__vector_quad* acc)
28 {
29  __builtin_mma_xxsetaccz(acc);
30 }
31 
32 template<typename DataMapper, typename Index, typename Packet, const Index accCols>
33 EIGEN_ALWAYS_INLINE void storeAccumulator(Index i, Index j, const DataMapper& data, const Packet& alpha, __vector_quad* acc)
34 {
36  __builtin_mma_disassemble_acc(&result.packet, acc);
37 
39  bload<DataMapper, Packet, Index, accCols, 0, ColMajor>(tRes, data, i, j);
40 
41  bscale<Packet>(tRes, result, alpha);
42 
43  data.template storePacketBlock<Packet, 4>(i, j, tRes);
44 }
45 
46 template<typename DataMapper, typename Index, typename Packet, typename Packetc, const Index accColsC, int N>
47 EIGEN_ALWAYS_INLINE void storeComplexAccumulator(Index i, Index j, const DataMapper& data, const Packet& alphaReal, const Packet& alphaImag, __vector_quad* accReal, __vector_quad* accImag)
48 {
49  PacketBlock<Packet, 4> resultReal, resultImag;
50  __builtin_mma_disassemble_acc(&resultReal.packet, accReal);
51  __builtin_mma_disassemble_acc(&resultImag.packet, accImag);
52 
54  bload<DataMapper, Packetc, Index, accColsC, N, ColMajor>(tRes, data, i, j);
55 
56  PacketBlock<Packet,4> taccReal, taccImag;
57  bscalec<Packet,4>(resultReal, resultImag, alphaReal, alphaImag, taccReal, taccImag);
58 
59  PacketBlock<Packetc, 4> acc1, acc2;
60  bcouple<Packet, Packetc>(taccReal, taccImag, tRes, acc1, acc2);
61 
62  data.template storePacketBlock<Packetc, 4>(i + N*accColsC, j, acc1);
63  data.template storePacketBlock<Packetc, 4>(i + (N+1)*accColsC, j, acc2);
64 }
65 
66 // Defaults to float32, since Eigen still supports C++03 we can't use default template arguments
67 template<typename LhsPacket, typename RhsPacket, bool NegativeAccumulate>
68 EIGEN_ALWAYS_INLINE void pgerMMA(__vector_quad* acc, const RhsPacket& a, const LhsPacket& b)
69 {
70  if(NegativeAccumulate)
71  {
72  __builtin_mma_xvf32gernp(acc, (__vector unsigned char)a, (__vector unsigned char)b);
73  } else {
74  __builtin_mma_xvf32gerpp(acc, (__vector unsigned char)a, (__vector unsigned char)b);
75  }
76 }
77 
78 template<typename LhsPacket, typename RhsPacket, bool NegativeAccumulate>
79 EIGEN_ALWAYS_INLINE void pgerMMA(__vector_quad* acc, const PacketBlock<Packet2d,2>& a, const Packet2d& b)
80 {
81  __vector_pair* a0 = (__vector_pair *)(&a.packet[0]);
82  if(NegativeAccumulate)
83  {
84  __builtin_mma_xvf64gernp(acc, *a0, (__vector unsigned char)b);
85  } else {
86  __builtin_mma_xvf64gerpp(acc, *a0, (__vector unsigned char)b);
87  }
88 }
89 
90 template<typename LhsPacket, typename RhsPacket, bool NegativeAccumulate>
91 EIGEN_ALWAYS_INLINE void pgerMMA(__vector_quad* acc, const __vector_pair& a, const Packet2d& b)
92 {
93  if(NegativeAccumulate)
94  {
95  __builtin_mma_xvf64gernp(acc, (__vector_pair)a, (__vector unsigned char)b);
96  } else {
97  __builtin_mma_xvf64gerpp(acc, (__vector_pair)a, (__vector unsigned char)b);
98  }
99 }
100 
101 template<typename LhsPacket, typename RhsPacket, bool NegativeAccumulate>
102 EIGEN_ALWAYS_INLINE void pgerMMA(__vector_quad*, const __vector_pair&, const Packet4f&)
103 {
104  // Just for compilation
105 }
106 
107 template<typename Scalar, typename Packet, typename RhsPacket, bool ConjugateLhs, bool ConjugateRhs, bool LhsIsReal, bool RhsIsReal>
108 EIGEN_ALWAYS_INLINE void pgercMMA(__vector_quad* accReal, __vector_quad* accImag, const Packet& lhsV, const Packet& lhsVi, const RhsPacket& rhsV, const RhsPacket& rhsVi)
109 {
110  pgerMMA<Packet, RhsPacket, false>(accReal, rhsV, lhsV);
111  if(LhsIsReal) {
112  pgerMMA<Packet, RhsPacket, ConjugateRhs>(accImag, rhsVi, lhsV);
113  } else {
114  if(!RhsIsReal) {
115  pgerMMA<Packet, RhsPacket, ConjugateLhs == ConjugateRhs>(accReal, rhsVi, lhsVi);
116  pgerMMA<Packet, RhsPacket, ConjugateRhs>(accImag, rhsVi, lhsV);
117  } else {
118  EIGEN_UNUSED_VARIABLE(rhsVi);
119  }
120  pgerMMA<Packet, RhsPacket, ConjugateLhs>(accImag, rhsV, lhsVi);
121  }
122 }
123 
124 // This is necessary because ploadRhs for double returns a pair of vectors when MMA is enabled.
125 template<typename Scalar, typename Packet>
127 {
128  rhsV = ploadRhs<Scalar, Packet>((const Scalar*)(rhs));
129 }
130 
131 template<>
132 EIGEN_ALWAYS_INLINE void ploadRhsMMA<double, PacketBlock<Packet2d, 2> >(const double* rhs, PacketBlock<Packet2d, 2>& rhsV)
133 {
134  rhsV.packet[0] = ploadRhs<double, Packet2d>((const double *)((Packet2d *)rhs ));
135  rhsV.packet[1] = ploadRhs<double, Packet2d>((const double *)(((Packet2d *)rhs) + 1));
136 }
137 
138 template<>
139 EIGEN_ALWAYS_INLINE void ploadRhsMMA<double, __vector_pair>(const double* rhs, __vector_pair& rhsV)
140 {
141 #if EIGEN_COMP_LLVM
142  __builtin_vsx_assemble_pair(&rhsV,
143  (__vector unsigned char)(ploadRhs<double, Packet2d>((const double *)(((Packet2d *)rhs) + 1))),
144  (__vector unsigned char)(ploadRhs<double, Packet2d>((const double *)((Packet2d *)rhs ))));
145 #else
146  __asm__ ("lxvp %x0,%1" : "=wa" (rhsV) : "Y" (*rhs));
147 #endif
148 }
149 
150 template<>
151 EIGEN_ALWAYS_INLINE void ploadRhsMMA(const float*, __vector_pair&)
152 {
153  // Just for compilation
154 }
155 
156 // PEEL_MMA loop factor.
157 #define PEEL_MMA 7
158 
159 #define MICRO_MMA_UNROLL(func) \
160  func(0) func(1) func(2) func(3) func(4) func(5) func(6) func(7)
161 
162 #define MICRO_MMA_LOAD_ONE(iter) \
163  if (unroll_factor > iter) { \
164  lhsV##iter = ploadLhs<Scalar, Packet>(lhs_ptr##iter); \
165  lhs_ptr##iter += accCols; \
166  } else { \
167  EIGEN_UNUSED_VARIABLE(lhsV##iter); \
168  }
169 
170 #define MICRO_MMA_WORK_ONE(iter, type, peel) \
171  if (unroll_factor > iter) { \
172  pgerMMA<Packet, type, false>(&accZero##iter, rhsV##peel, lhsV##iter); \
173  }
174 
175 #define MICRO_MMA_TYPE_PEEL(func, func2, type, peel) \
176  if (PEEL_MMA > peel) { \
177  Packet lhsV0, lhsV1, lhsV2, lhsV3, lhsV4, lhsV5, lhsV6, lhsV7; \
178  ploadRhsMMA<Scalar, type>(rhs_ptr + (accRows * peel), rhsV##peel); \
179  MICRO_MMA_UNROLL(func2); \
180  func(0,type,peel) func(1,type,peel) func(2,type,peel) func(3,type,peel) \
181  func(4,type,peel) func(5,type,peel) func(6,type,peel) func(7,type,peel) \
182  } else { \
183  EIGEN_UNUSED_VARIABLE(rhsV##peel); \
184  }
185 
186 #define MICRO_MMA_UNROLL_TYPE_PEEL(func, func2, type) \
187  type rhsV0, rhsV1, rhsV2, rhsV3, rhsV4, rhsV5, rhsV6, rhsV7, rhsV8, rhsV9; \
188  MICRO_MMA_TYPE_PEEL(func,func2,type,0); MICRO_MMA_TYPE_PEEL(func,func2,type,1); \
189  MICRO_MMA_TYPE_PEEL(func,func2,type,2); MICRO_MMA_TYPE_PEEL(func,func2,type,3); \
190  MICRO_MMA_TYPE_PEEL(func,func2,type,4); MICRO_MMA_TYPE_PEEL(func,func2,type,5); \
191  MICRO_MMA_TYPE_PEEL(func,func2,type,6); MICRO_MMA_TYPE_PEEL(func,func2,type,7); \
192  MICRO_MMA_TYPE_PEEL(func,func2,type,8); MICRO_MMA_TYPE_PEEL(func,func2,type,9);
193 
194 #define MICRO_MMA_UNROLL_TYPE_ONE(func, func2, type) \
195  type rhsV0; \
196  MICRO_MMA_TYPE_PEEL(func,func2,type,0);
197 
198 #define MICRO_MMA_ONE_PEEL \
199  if (sizeof(Scalar) == sizeof(float)) { \
200  MICRO_MMA_UNROLL_TYPE_PEEL(MICRO_MMA_WORK_ONE, MICRO_MMA_LOAD_ONE, RhsPacket); \
201  } else { \
202  MICRO_MMA_UNROLL_TYPE_PEEL(MICRO_MMA_WORK_ONE, MICRO_MMA_LOAD_ONE, __vector_pair); \
203  } \
204  rhs_ptr += (accRows * PEEL_MMA);
205 
206 #define MICRO_MMA_ONE \
207  if (sizeof(Scalar) == sizeof(float)) { \
208  MICRO_MMA_UNROLL_TYPE_ONE(MICRO_MMA_WORK_ONE, MICRO_MMA_LOAD_ONE, RhsPacket); \
209  } else { \
210  MICRO_MMA_UNROLL_TYPE_ONE(MICRO_MMA_WORK_ONE, MICRO_MMA_LOAD_ONE, __vector_pair); \
211  } \
212  rhs_ptr += accRows;
213 
214 #define MICRO_MMA_DST_PTR_ONE(iter) \
215  if (unroll_factor > iter) { \
216  bsetzeroMMA<Scalar, Packet>(&accZero##iter); \
217  } else { \
218  EIGEN_UNUSED_VARIABLE(accZero##iter); \
219  }
220 
221 #define MICRO_MMA_DST_PTR MICRO_MMA_UNROLL(MICRO_MMA_DST_PTR_ONE)
222 
223 #define MICRO_MMA_SRC_PTR_ONE(iter) \
224  if (unroll_factor > iter) { \
225  lhs_ptr##iter = lhs_base + ( (row/accCols) + iter )*strideA*accCols + accCols*offsetA; \
226  } else { \
227  EIGEN_UNUSED_VARIABLE(lhs_ptr##iter); \
228  }
229 
230 #define MICRO_MMA_SRC_PTR MICRO_MMA_UNROLL(MICRO_MMA_SRC_PTR_ONE)
231 
232 #define MICRO_MMA_PREFETCH_ONE(iter) \
233  if (unroll_factor > iter) { \
234  EIGEN_POWER_PREFETCH(lhs_ptr##iter); \
235  }
236 
237 #define MICRO_MMA_PREFETCH MICRO_MMA_UNROLL(MICRO_MMA_PREFETCH_ONE)
238 
239 #define MICRO_MMA_STORE_ONE(iter) \
240  if (unroll_factor > iter) { \
241  storeAccumulator<DataMapper, Index, Packet, accCols>(row + iter*accCols, col, res, pAlpha, &accZero##iter); \
242  }
243 
244 #define MICRO_MMA_STORE MICRO_MMA_UNROLL(MICRO_MMA_STORE_ONE)
245 
246 template<int unroll_factor, typename Scalar, typename Packet, typename RhsPacket, typename DataMapper, typename Index, const Index accRows, const Index accCols>
248  const DataMapper& res,
249  const Scalar* lhs_base,
250  const Scalar* rhs_base,
251  Index depth,
252  Index strideA,
253  Index offsetA,
254  Index& row,
255  Index col,
256  const Packet& pAlpha)
257 {
258  const Scalar* rhs_ptr = rhs_base;
259  const Scalar* lhs_ptr0 = NULL, * lhs_ptr1 = NULL, * lhs_ptr2 = NULL, * lhs_ptr3 = NULL, * lhs_ptr4 = NULL, * lhs_ptr5 = NULL, * lhs_ptr6 = NULL, * lhs_ptr7 = NULL;
260  __vector_quad accZero0, accZero1, accZero2, accZero3, accZero4, accZero5, accZero6, accZero7;
261 
264 
265  Index k = 0;
266  for(; k + PEEL_MMA <= depth; k+= PEEL_MMA)
267  {
268  EIGEN_POWER_PREFETCH(rhs_ptr);
271  }
272  for(; k < depth; k++)
273  {
275  }
277 
278  row += unroll_factor*accCols;
279 }
280 
281 template<typename Scalar, typename Index, typename Packet, typename RhsPacket, typename DataMapper, const Index accRows, const Index accCols>
282 void gemmMMA(const DataMapper& res, const Scalar* blockA, const Scalar* blockB, Index rows, Index depth, Index cols, Scalar alpha, Index strideA, Index strideB, Index offsetA, Index offsetB)
283 {
284  const Index remaining_rows = rows % accCols;
285  const Index remaining_cols = cols % accRows;
286 
287  if( strideA == -1 ) strideA = depth;
288  if( strideB == -1 ) strideB = depth;
289 
290  const Packet pAlpha = pset1<Packet>(alpha);
291  const Packet pMask = bmask<Packet>((const int)(remaining_rows));
292 
293  Index col = 0;
294  for(; col + accRows <= cols; col += accRows)
295  {
296  const Scalar* rhs_base = blockB + col*strideB + accRows*offsetB;
297  const Scalar* lhs_base = blockA;
298 
299  Index row = 0;
300 #define MAX_MMA_UNROLL 7
301  while(row + MAX_MMA_UNROLL*accCols <= rows) {
302  gemm_unrolled_MMA_iteration<MAX_MMA_UNROLL, Scalar, Packet, RhsPacket, DataMapper, Index, accRows, accCols>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, col, pAlpha);
303  }
304  switch( (rows-row)/accCols ) {
305 #if MAX_MMA_UNROLL > 7
306  case 7:
307  gemm_unrolled_MMA_iteration<7, Scalar, Packet, RhsPacket, DataMapper, Index, accRows, accCols>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, col, pAlpha);
308  break;
309 #endif
310 #if MAX_MMA_UNROLL > 6
311  case 6:
312  gemm_unrolled_MMA_iteration<6, Scalar, Packet, RhsPacket, DataMapper, Index, accRows, accCols>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, col, pAlpha);
313  break;
314 #endif
315 #if MAX_MMA_UNROLL > 5
316  case 5:
317  gemm_unrolled_MMA_iteration<5, Scalar, Packet, RhsPacket, DataMapper, Index, accRows, accCols>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, col, pAlpha);
318  break;
319 #endif
320 #if MAX_MMA_UNROLL > 4
321  case 4:
322  gemm_unrolled_MMA_iteration<4, Scalar, Packet, RhsPacket, DataMapper, Index, accRows, accCols>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, col, pAlpha);
323  break;
324 #endif
325 #if MAX_MMA_UNROLL > 3
326  case 3:
327  gemm_unrolled_MMA_iteration<3, Scalar, Packet, RhsPacket, DataMapper, Index, accRows, accCols>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, col, pAlpha);
328  break;
329 #endif
330 #if MAX_MMA_UNROLL > 2
331  case 2:
332  gemm_unrolled_MMA_iteration<2, Scalar, Packet, RhsPacket, DataMapper, Index, accRows, accCols>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, col, pAlpha);
333  break;
334 #endif
335 #if MAX_MMA_UNROLL > 1
336  case 1:
337  gemm_unrolled_MMA_iteration<1, Scalar, Packet, RhsPacket, DataMapper, Index, accRows, accCols>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, col, pAlpha);
338  break;
339 #endif
340  default:
341  break;
342  }
343 #undef MAX_MMA_UNROLL
344 
345  if(remaining_rows > 0)
346  {
347  gemm_extra_row<Scalar, Packet, DataMapper, Index, accRows, accCols>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, col, rows, cols, remaining_rows, pAlpha, pMask);
348  }
349  }
350 
351  if(remaining_cols > 0)
352  {
353  const Scalar* rhs_base = blockB + col*strideB + remaining_cols*offsetB;
354  const Scalar* lhs_base = blockA;
355 
356  for(; col < cols; col++)
357  {
358  Index row = 0;
359 
360  gemm_unrolled_col<Scalar, Packet, DataMapper, Index, accCols>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, rows, col, remaining_cols, pAlpha);
361 
362  if (remaining_rows > 0)
363  {
364  gemm_extra_col<Scalar, Packet, DataMapper, Index, accRows>(res, lhs_base, rhs_base, depth, strideA, offsetA, row, col, remaining_rows, remaining_cols, pAlpha);
365  }
366  rhs_base++;
367  }
368  }
369 }
370 
371 #define accColsC (accCols / 2)
372 #define advanceRows ((LhsIsReal) ? 1 : 2)
373 #define advanceCols ((RhsIsReal) ? 1 : 2)
374 
375 // PEEL_COMPLEX_MMA loop factor.
376 #define PEEL_COMPLEX_MMA 7
377 
378 #define MICRO_COMPLEX_MMA_UNROLL(func) \
379  func(0) func(1) func(2) func(3) func(4)
380 
381 #define MICRO_COMPLEX_MMA_LOAD_ONE(iter) \
382  if (unroll_factor > iter) { \
383  lhsV##iter = ploadLhs<Scalar, Packet>(lhs_ptr_real##iter); \
384  lhs_ptr_real##iter += accCols; \
385  if(!LhsIsReal) { \
386  lhsVi##iter = ploadLhs<Scalar, Packet>(lhs_ptr_imag##iter); \
387  lhs_ptr_imag##iter += accCols; \
388  } else { \
389  EIGEN_UNUSED_VARIABLE(lhsVi##iter); \
390  } \
391  } else { \
392  EIGEN_UNUSED_VARIABLE(lhsV##iter); \
393  EIGEN_UNUSED_VARIABLE(lhsVi##iter); \
394  }
395 
396 #define MICRO_COMPLEX_MMA_WORK_ONE(iter, type, peel) \
397  if (unroll_factor > iter) { \
398  pgercMMA<Scalar, Packet, type, ConjugateLhs, ConjugateRhs, LhsIsReal, RhsIsReal>(&accReal##iter, &accImag##iter, lhsV##iter, lhsVi##iter, rhsV##peel, rhsVi##peel); \
399  }
400 
401 #define MICRO_COMPLEX_MMA_TYPE_PEEL(func, func2, type, peel) \
402  if (PEEL_COMPLEX_MMA > peel) { \
403  Packet lhsV0, lhsV1, lhsV2, lhsV3, lhsV4; \
404  Packet lhsVi0, lhsVi1, lhsVi2, lhsVi3, lhsVi4; \
405  ploadRhsMMA<Scalar, type>(rhs_ptr_real + (accRows * peel), rhsV##peel); \
406  if(!RhsIsReal) { \
407  ploadRhsMMA<Scalar, type>(rhs_ptr_imag + (accRows * peel), rhsVi##peel); \
408  } else { \
409  EIGEN_UNUSED_VARIABLE(rhsVi##peel); \
410  } \
411  MICRO_COMPLEX_MMA_UNROLL(func2); \
412  func(0,type,peel) func(1,type,peel) func(2,type,peel) func(3,type,peel) func(4,type,peel) \
413  } else { \
414  EIGEN_UNUSED_VARIABLE(rhsV##peel); \
415  EIGEN_UNUSED_VARIABLE(rhsVi##peel); \
416  }
417 
418 #define MICRO_COMPLEX_MMA_UNROLL_TYPE_PEEL(func, func2, type) \
419  type rhsV0, rhsV1, rhsV2, rhsV3, rhsV4, rhsV5, rhsV6, rhsV7, rhsV8, rhsV9; \
420  type rhsVi0, rhsVi1, rhsVi2, rhsVi3, rhsVi4, rhsVi5, rhsVi6, rhsVi7, rhsVi8, rhsVi9; \
421  MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,0); MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,1); \
422  MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,2); MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,3); \
423  MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,4); MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,5); \
424  MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,6); MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,7); \
425  MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,8); MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,9);
426 
427 #define MICRO_COMPLEX_MMA_UNROLL_TYPE_ONE(func, func2, type) \
428  type rhsV0, rhsVi0; \
429  MICRO_COMPLEX_MMA_TYPE_PEEL(func,func2,type,0);
430 
431 #define MICRO_COMPLEX_MMA_ONE_PEEL \
432  if (sizeof(Scalar) == sizeof(float)) { \
433  MICRO_COMPLEX_MMA_UNROLL_TYPE_PEEL(MICRO_COMPLEX_MMA_WORK_ONE, MICRO_COMPLEX_MMA_LOAD_ONE, RhsPacket); \
434  } else { \
435  MICRO_COMPLEX_MMA_UNROLL_TYPE_PEEL(MICRO_COMPLEX_MMA_WORK_ONE, MICRO_COMPLEX_MMA_LOAD_ONE, __vector_pair); \
436  } \
437  rhs_ptr_real += (accRows * PEEL_COMPLEX_MMA); \
438  if(!RhsIsReal) rhs_ptr_imag += (accRows * PEEL_COMPLEX_MMA);
439 
440 #define MICRO_COMPLEX_MMA_ONE \
441  if (sizeof(Scalar) == sizeof(float)) { \
442  MICRO_COMPLEX_MMA_UNROLL_TYPE_ONE(MICRO_COMPLEX_MMA_WORK_ONE, MICRO_COMPLEX_MMA_LOAD_ONE, RhsPacket); \
443  } else { \
444  MICRO_COMPLEX_MMA_UNROLL_TYPE_ONE(MICRO_COMPLEX_MMA_WORK_ONE, MICRO_COMPLEX_MMA_LOAD_ONE, __vector_pair); \
445  } \
446  rhs_ptr_real += accRows; \
447  if(!RhsIsReal) rhs_ptr_imag += accRows;
448 
449 #define MICRO_COMPLEX_MMA_DST_PTR_ONE(iter) \
450  if (unroll_factor > iter) { \
451  bsetzeroMMA<Scalar, Packet>(&accReal##iter); \
452  bsetzeroMMA<Scalar, Packet>(&accImag##iter); \
453  } else { \
454  EIGEN_UNUSED_VARIABLE(accReal##iter); \
455  EIGEN_UNUSED_VARIABLE(accImag##iter); \
456  }
457 
458 #define MICRO_COMPLEX_MMA_DST_PTR MICRO_COMPLEX_MMA_UNROLL(MICRO_COMPLEX_MMA_DST_PTR_ONE)
459 
460 #define MICRO_COMPLEX_MMA_SRC_PTR_ONE(iter) \
461  if (unroll_factor > iter) { \
462  lhs_ptr_real##iter = lhs_base + ( ((advanceRows*row)/accCols) + iter*advanceRows )*strideA*accCols + accCols*offsetA; \
463  if(!LhsIsReal) { \
464  lhs_ptr_imag##iter = lhs_ptr_real##iter + accCols*strideA; \
465  } else { \
466  EIGEN_UNUSED_VARIABLE(lhs_ptr_imag##iter); \
467  } \
468  } else { \
469  EIGEN_UNUSED_VARIABLE(lhs_ptr_real##iter); \
470  EIGEN_UNUSED_VARIABLE(lhs_ptr_imag##iter); \
471  }
472 
473 #define MICRO_COMPLEX_MMA_SRC_PTR MICRO_COMPLEX_MMA_UNROLL(MICRO_COMPLEX_MMA_SRC_PTR_ONE)
474 
475 #define MICRO_COMPLEX_MMA_PREFETCH_ONE(iter) \
476  if (unroll_factor > iter) { \
477  EIGEN_POWER_PREFETCH(lhs_ptr_real##iter); \
478  if(!LhsIsReal) { \
479  EIGEN_POWER_PREFETCH(lhs_ptr_imag##iter); \
480  } \
481  }
482 
483 #define MICRO_COMPLEX_MMA_PREFETCH MICRO_COMPLEX_MMA_UNROLL(MICRO_COMPLEX_MMA_PREFETCH_ONE)
484 
485 #define MICRO_COMPLEX_MMA_STORE_ONE(iter) \
486  if (unroll_factor > iter) { \
487  storeComplexAccumulator<DataMapper, Index, Packet, Packetc, accColsC, 0>(row + iter*accCols, col, res, pAlphaReal, pAlphaImag, &accReal##iter, &accImag##iter); \
488  }
489 
490 #define MICRO_COMPLEX_MMA_STORE MICRO_COMPLEX_MMA_UNROLL(MICRO_COMPLEX_MMA_STORE_ONE)
491 
492 template<int unroll_factor, typename Scalar, typename Packet, typename Packetc, typename RhsPacket, typename DataMapper, typename Index, const Index accRows, const Index accCols, bool ConjugateLhs, bool ConjugateRhs, bool LhsIsReal, bool RhsIsReal>
494  const DataMapper& res,
495  const Scalar* lhs_base,
496  const Scalar* rhs_base,
497  Index depth,
498  Index strideA,
499  Index offsetA,
500  Index strideB,
501  Index& row,
502  Index col,
503  const Packet& pAlphaReal,
504  const Packet& pAlphaImag)
505 {
506  const Scalar* rhs_ptr_real = rhs_base;
507  const Scalar* rhs_ptr_imag;
508  if(!RhsIsReal) {
509  rhs_ptr_imag = rhs_base + accRows*strideB;
510  } else {
511  EIGEN_UNUSED_VARIABLE(rhs_ptr_imag);
512  }
513  const Scalar* lhs_ptr_real0 = NULL, * lhs_ptr_imag0 = NULL, * lhs_ptr_real1 = NULL, * lhs_ptr_imag1 = NULL;
514  const Scalar* lhs_ptr_real2 = NULL, * lhs_ptr_imag2 = NULL, * lhs_ptr_real3 = NULL, * lhs_ptr_imag3 = NULL;
515  const Scalar* lhs_ptr_real4 = NULL, * lhs_ptr_imag4 = NULL;
516  __vector_quad accReal0, accImag0, accReal1, accImag1, accReal2, accImag2, accReal3, accImag3, accReal4, accImag4;
517 
520 
521  Index k = 0;
522  for(; k + PEEL_COMPLEX_MMA <= depth; k+= PEEL_COMPLEX_MMA)
523  {
524  EIGEN_POWER_PREFETCH(rhs_ptr_real);
525  if(!RhsIsReal) {
526  EIGEN_POWER_PREFETCH(rhs_ptr_imag);
527  }
530  }
531  for(; k < depth; k++)
532  {
534  }
536 
537  row += unroll_factor*accCols;
538 }
539 
540 template<typename LhsScalar, typename RhsScalar, typename Scalarc, typename Scalar, typename Index, typename Packet, typename Packetc, typename RhsPacket, typename DataMapper, const Index accRows, const Index accCols, bool ConjugateLhs, bool ConjugateRhs, bool LhsIsReal, bool RhsIsReal>
541 void gemm_complexMMA(const DataMapper& res, const LhsScalar* blockAc, const RhsScalar* blockBc, Index rows, Index depth, Index cols, Scalarc alpha, Index strideA, Index strideB, Index offsetA, Index offsetB)
542 {
543  const Index remaining_rows = rows % accCols;
544  const Index remaining_cols = cols % accRows;
545 
546  if( strideA == -1 ) strideA = depth;
547  if( strideB == -1 ) strideB = depth;
548 
549  const Packet pAlphaReal = pset1<Packet>(alpha.real());
550  const Packet pAlphaImag = pset1<Packet>(alpha.imag());
551  const Packet pMask = bmask<Packet>((const int)(remaining_rows));
552 
553  const Scalar* blockA = (Scalar *) blockAc;
554  const Scalar* blockB = (Scalar *) blockBc;
555 
556  Index col = 0;
557  for(; col + accRows <= cols; col += accRows)
558  {
559  const Scalar* rhs_base = blockB + advanceCols*col*strideB + accRows*offsetB;
560  const Scalar* lhs_base = blockA;
561  Index row = 0;
562 
563 #define MAX_COMPLEX_MMA_UNROLL 4
564  while(row + MAX_COMPLEX_MMA_UNROLL*accCols <= rows) {
565  gemm_complex_unrolled_MMA_iteration<MAX_COMPLEX_MMA_UNROLL, Scalar, Packet, Packetc, RhsPacket, DataMapper, Index, accRows, accCols, ConjugateLhs, ConjugateRhs, LhsIsReal, RhsIsReal>(res, lhs_base, rhs_base, depth, strideA, offsetA, strideB, row, col, pAlphaReal, pAlphaImag);
566  }
567  switch( (rows-row)/accCols ) {
568 #if MAX_COMPLEX_MMA_UNROLL > 4
569  case 4:
570  gemm_complex_unrolled_MMA_iteration<4, Scalar, Packet, Packetc, RhsPacket, DataMapper, Index, accRows, accCols, ConjugateLhs, ConjugateRhs, LhsIsReal, RhsIsReal>(res, lhs_base, rhs_base, depth, strideA, offsetA, strideB, row, col, pAlphaReal, pAlphaImag);
571  break;
572 #endif
573 #if MAX_COMPLEX_MMA_UNROLL > 3
574  case 3:
575  gemm_complex_unrolled_MMA_iteration<3, Scalar, Packet, Packetc, RhsPacket, DataMapper, Index, accRows, accCols, ConjugateLhs, ConjugateRhs, LhsIsReal, RhsIsReal>(res, lhs_base, rhs_base, depth, strideA, offsetA, strideB, row, col, pAlphaReal, pAlphaImag);
576  break;
577 #endif
578 #if MAX_COMPLEX_MMA_UNROLL > 2
579  case 2:
580  gemm_complex_unrolled_MMA_iteration<2, Scalar, Packet, Packetc, RhsPacket, DataMapper, Index, accRows, accCols, ConjugateLhs, ConjugateRhs, LhsIsReal, RhsIsReal>(res, lhs_base, rhs_base, depth, strideA, offsetA, strideB, row, col, pAlphaReal, pAlphaImag);
581  break;
582 #endif
583 #if MAX_COMPLEX_MMA_UNROLL > 1
584  case 1:
585  gemm_complex_unrolled_MMA_iteration<1, Scalar, Packet, Packetc, RhsPacket, DataMapper, Index, accRows, accCols, ConjugateLhs, ConjugateRhs, LhsIsReal, RhsIsReal>(res, lhs_base, rhs_base, depth, strideA, offsetA, strideB, row, col, pAlphaReal, pAlphaImag);
586  break;
587 #endif
588  default:
589  break;
590  }
591 #undef MAX_COMPLEX_MMA_UNROLL
592 
593  if(remaining_rows > 0)
594  {
595  gemm_complex_extra_row<Scalar, Packet, Packetc, DataMapper, Index, accRows, accCols, ConjugateLhs, ConjugateRhs, LhsIsReal, RhsIsReal>(res, lhs_base, rhs_base, depth, strideA, offsetA, strideB, row, col, rows, cols, remaining_rows, pAlphaReal, pAlphaImag, pMask);
596  }
597  }
598 
599  if(remaining_cols > 0)
600  {
601  const Scalar* rhs_base = blockB + advanceCols*col*strideB + remaining_cols*offsetB;
602  const Scalar* lhs_base = blockA;
603 
604  for(; col < cols; col++)
605  {
606  Index row = 0;
607 
608  gemm_complex_unrolled_col<Scalar, Packet, Packetc, DataMapper, Index, accCols, ConjugateLhs, ConjugateRhs, LhsIsReal, RhsIsReal>(res, lhs_base, rhs_base, depth, strideA, offsetA, strideB, row, rows, col, remaining_cols, pAlphaReal, pAlphaImag);
609 
610  if (remaining_rows > 0)
611  {
612  gemm_complex_extra_col<Scalar, Packet, Packetc, DataMapper, Index, accRows, accCols, ConjugateLhs, ConjugateRhs, LhsIsReal, RhsIsReal>(res, lhs_base, rhs_base, depth, strideA, offsetA, strideB, row, col, remaining_rows, remaining_cols, pAlphaReal, pAlphaImag);
613  }
614  rhs_base++;
615  }
616  }
617 }
618 
619 #undef accColsC
620 #undef advanceRows
621 #undef advanceCols
622 
623 #pragma GCC reset_options
624 } // end namespace internal
625 
626 } // end namespace Eigen
627 
628 #endif // EIGEN_MATRIX_PRODUCT_MMA_ALTIVEC_H
629 
MICRO_COMPLEX_MMA_SRC_PTR
#define MICRO_COMPLEX_MMA_SRC_PTR
Definition: MatrixProductMMA.h:473
MICRO_MMA_ONE_PEEL
#define MICRO_MMA_ONE_PEEL
Definition: MatrixProductMMA.h:198
Eigen
Namespace containing all symbols from the Eigen library.
Definition: jet.h:637
col
m col(1)
Eigen::internal::bsetzeroMMA
EIGEN_ALWAYS_INLINE void bsetzeroMMA(__vector_quad *acc)
Definition: MatrixProductMMA.h:27
Eigen::internal::Packet4f
__vector float Packet4f
Definition: AltiVec/PacketMath.h:30
MICRO_MMA_STORE
#define MICRO_MMA_STORE
Definition: MatrixProductMMA.h:244
alpha
RealScalar alpha
Definition: level1_cplx_impl.h:147
MICRO_MMA_ONE
#define MICRO_MMA_ONE
Definition: MatrixProductMMA.h:206
MICRO_MMA_PREFETCH
#define MICRO_MMA_PREFETCH
Definition: MatrixProductMMA.h:237
Eigen::internal::PacketBlock
Definition: GenericPacketMath.h:1014
MICRO_COMPLEX_MMA_PREFETCH
#define MICRO_COMPLEX_MMA_PREFETCH
Definition: MatrixProductMMA.h:483
b
Scalar * b
Definition: benchVecAdd.cpp:17
MAX_COMPLEX_MMA_UNROLL
#define MAX_COMPLEX_MMA_UNROLL
PEEL_MMA
#define PEEL_MMA
Definition: MatrixProductMMA.h:157
Eigen::internal::gemm_unrolled_MMA_iteration
EIGEN_STRONG_INLINE void gemm_unrolled_MMA_iteration(const DataMapper &res, const Scalar *lhs_base, const Scalar *rhs_base, Index depth, Index strideA, Index offsetA, Index &row, Index col, const Packet &pAlpha)
Definition: MatrixProductMMA.h:247
MICRO_COMPLEX_MMA_ONE
#define MICRO_COMPLEX_MMA_ONE
Definition: MatrixProductMMA.h:440
MICRO_COMPLEX_MMA_STORE
#define MICRO_COMPLEX_MMA_STORE
Definition: MatrixProductMMA.h:490
Eigen::internal::PacketBlock::packet
Packet packet[N]
Definition: GenericPacketMath.h:1018
accColsC
#define accColsC
Definition: MatrixProductMMA.h:371
res
cout<< "Here is the matrix m:"<< endl<< m<< endl;Matrix< ptrdiff_t, 3, 1 > res
Definition: PartialRedux_count.cpp:3
MICRO_COMPLEX_MMA_DST_PTR
#define MICRO_COMPLEX_MMA_DST_PTR
Definition: MatrixProductMMA.h:458
result
Values result
Definition: OdometryOptimize.cpp:8
rows
int rows
Definition: Tutorial_commainit_02.cpp:1
Eigen::internal::ploadRhsMMA
EIGEN_ALWAYS_INLINE void ploadRhsMMA(const Scalar *rhs, Packet &rhsV)
Definition: MatrixProductMMA.h:126
Eigen::internal::gemm_complexMMA
void gemm_complexMMA(const DataMapper &res, const LhsScalar *blockAc, const RhsScalar *blockBc, Index rows, Index depth, Index cols, Scalarc alpha, Index strideA, Index strideB, Index offsetA, Index offsetB)
Definition: MatrixProductMMA.h:541
Eigen::internal::storeComplexAccumulator
EIGEN_ALWAYS_INLINE void storeComplexAccumulator(Index i, Index j, const DataMapper &data, const Packet &alphaReal, const Packet &alphaImag, __vector_quad *accReal, __vector_quad *accImag)
Definition: MatrixProductMMA.h:47
data
int data[]
Definition: Map_placement_new.cpp:1
MICRO_MMA_DST_PTR
#define MICRO_MMA_DST_PTR
Definition: MatrixProductMMA.h:221
EIGEN_UNUSED_VARIABLE
#define EIGEN_UNUSED_VARIABLE(var)
Definition: Macros.h:1076
j
std::ptrdiff_t j
Definition: tut_arithmetic_redux_minmax.cpp:2
Eigen::internal::ploadRhsMMA< double, __vector_pair >
EIGEN_ALWAYS_INLINE void ploadRhsMMA< double, __vector_pair >(const double *rhs, __vector_pair &rhsV)
Definition: MatrixProductMMA.h:139
EIGEN_STRONG_INLINE
#define EIGEN_STRONG_INLINE
Definition: Macros.h:917
Eigen::internal::Packet
Definition: ZVector/PacketMath.h:47
EIGEN_ALWAYS_INLINE
#define EIGEN_ALWAYS_INLINE
Definition: Macros.h:932
advanceCols
#define advanceCols
Definition: MatrixProductMMA.h:373
EIGEN_POWER_PREFETCH
#define EIGEN_POWER_PREFETCH(p)
Definition: MatrixProductCommon.h:5
Eigen::internal::gemmMMA
void gemmMMA(const DataMapper &res, const Scalar *blockA, const Scalar *blockB, Index rows, Index depth, Index cols, Scalar alpha, Index strideA, Index strideB, Index offsetA, Index offsetB)
Definition: MatrixProductMMA.h:282
MICRO_COMPLEX_MMA_ONE_PEEL
#define MICRO_COMPLEX_MMA_ONE_PEEL
Definition: MatrixProductMMA.h:431
a
ArrayXXi a
Definition: Array_initializer_list_23_cxx11.cpp:1
MAX_MMA_UNROLL
#define MAX_MMA_UNROLL
MICRO_MMA_SRC_PTR
#define MICRO_MMA_SRC_PTR
Definition: MatrixProductMMA.h:230
Eigen::internal::pgercMMA
EIGEN_ALWAYS_INLINE void pgercMMA(__vector_quad *accReal, __vector_quad *accImag, const Packet &lhsV, const Packet &lhsVi, const RhsPacket &rhsV, const RhsPacket &rhsVi)
Definition: MatrixProductMMA.h:108
Eigen::internal::Packet2d
v2f64 Packet2d
Definition: MSA/PacketMath.h:820
row
m row(1)
Eigen::internal::pgerMMA
EIGEN_ALWAYS_INLINE void pgerMMA(__vector_quad *acc, const RhsPacket &a, const LhsPacket &b)
Definition: MatrixProductMMA.h:68
N
#define N
Definition: igam.h:9
internal
Definition: BandTriangularSolver.h:13
NULL
#define NULL
Definition: ccolamd.c:609
cols
int cols
Definition: Tutorial_commainit_02.cpp:1
PEEL_COMPLEX_MMA
#define PEEL_COMPLEX_MMA
Definition: MatrixProductMMA.h:376
Eigen::internal::gemm_complex_unrolled_MMA_iteration
EIGEN_STRONG_INLINE void gemm_complex_unrolled_MMA_iteration(const DataMapper &res, const Scalar *lhs_base, const Scalar *rhs_base, Index depth, Index strideA, Index offsetA, Index strideB, Index &row, Index col, const Packet &pAlphaReal, const Packet &pAlphaImag)
Definition: MatrixProductMMA.h:493
depth
static double depth
Definition: testSphericalCamera.cpp:64
i
int i
Definition: BiCGSTAB_step_by_step.cpp:9
Scalar
SCALAR Scalar
Definition: bench_gemm.cpp:46
Eigen::Index
EIGEN_DEFAULT_DENSE_INDEX_TYPE Index
The Index type as used for the API.
Definition: Meta.h:74
Eigen::internal::storeAccumulator
EIGEN_ALWAYS_INLINE void storeAccumulator(Index i, Index j, const DataMapper &data, const Packet &alpha, __vector_quad *acc)
Definition: MatrixProductMMA.h:33


gtsam
Author(s):
autogenerated on Sat Nov 16 2024 04:03:09