gonum.org/v1/gonum@v0.14.0/internal/asm/f64/gemvT_amd64.s (about)

     1  // Copyright ©2017 The Gonum Authors. All rights reserved.
     2  // Use of this source code is governed by a BSD-style
     3  // license that can be found in the LICENSE file.
     4  
     5  // +build !noasm,!gccgo,!safe
     6  
     7  #include "textflag.h"
     8  
     9  #define SIZE 8
    10  
    11  #define M_DIM n+8(FP)
    12  #define M CX
    13  #define N_DIM m+0(FP)
    14  #define N BX
    15  
    16  #define TMP1 R14
    17  #define TMP2 R15
    18  
    19  #define X_PTR SI
    20  #define X x_base+56(FP)
    21  #define Y_PTR DX
    22  #define Y y_base+96(FP)
    23  #define A_ROW AX
    24  #define A_PTR DI
    25  
    26  #define INC_X R8
    27  #define INC3_X R9
    28  
    29  #define INC_Y R10
    30  #define INC3_Y R11
    31  
    32  #define LDA R12
    33  #define LDA3 R13
    34  
    35  #define ALPHA X15
    36  #define BETA X14
    37  
    38  #define INIT4 \
    39  	MOVDDUP (X_PTR), X8            \
    40  	MOVDDUP (X_PTR)(INC_X*1), X9   \
    41  	MOVDDUP (X_PTR)(INC_X*2), X10  \
    42  	MOVDDUP (X_PTR)(INC3_X*1), X11 \
    43  	MULPD   ALPHA, X8              \
    44  	MULPD   ALPHA, X9              \
    45  	MULPD   ALPHA, X10             \
    46  	MULPD   ALPHA, X11
    47  
    48  #define INIT2 \
    49  	MOVDDUP (X_PTR), X8          \
    50  	MOVDDUP (X_PTR)(INC_X*1), X9 \
    51  	MULPD   ALPHA, X8            \
    52  	MULPD   ALPHA, X9
    53  
    54  #define INIT1 \
    55  	MOVDDUP (X_PTR), X8 \
    56  	MULPD   ALPHA, X8
    57  
    58  #define KERNEL_LOAD4 \
    59  	MOVUPS (Y_PTR), X0       \
    60  	MOVUPS 2*SIZE(Y_PTR), X1
    61  
    62  #define KERNEL_LOAD2 \
    63  	MOVUPS (Y_PTR), X0
    64  
    65  #define KERNEL_LOAD4_INC \
    66  	MOVSD  (Y_PTR), X0           \
    67  	MOVHPD (Y_PTR)(INC_Y*1), X0  \
    68  	MOVSD  (Y_PTR)(INC_Y*2), X1  \
    69  	MOVHPD (Y_PTR)(INC3_Y*1), X1
    70  
    71  #define KERNEL_LOAD2_INC \
    72  	MOVSD  (Y_PTR), X0          \
    73  	MOVHPD (Y_PTR)(INC_Y*1), X0
    74  
    75  #define KERNEL_4x4 \
    76  	MOVUPS (A_PTR), X4               \
    77  	MOVUPS 2*SIZE(A_PTR), X5         \
    78  	MOVUPS (A_PTR)(LDA*1), X6        \
    79  	MOVUPS 2*SIZE(A_PTR)(LDA*1), X7  \
    80  	MULPD  X8, X4                    \
    81  	MULPD  X8, X5                    \
    82  	MULPD  X9, X6                    \
    83  	MULPD  X9, X7                    \
    84  	ADDPD  X4, X0                    \
    85  	ADDPD  X5, X1                    \
    86  	ADDPD  X6, X0                    \
    87  	ADDPD  X7, X1                    \
    88  	MOVUPS (A_PTR)(LDA*2), X4        \
    89  	MOVUPS 2*SIZE(A_PTR)(LDA*2), X5  \
    90  	MOVUPS (A_PTR)(LDA3*1), X6       \
    91  	MOVUPS 2*SIZE(A_PTR)(LDA3*1), X7 \
    92  	MULPD  X10, X4                   \
    93  	MULPD  X10, X5                   \
    94  	MULPD  X11, X6                   \
    95  	MULPD  X11, X7                   \
    96  	ADDPD  X4, X0                    \
    97  	ADDPD  X5, X1                    \
    98  	ADDPD  X6, X0                    \
    99  	ADDPD  X7, X1                    \
   100  	ADDQ   $4*SIZE, A_PTR
   101  
   102  #define KERNEL_4x2 \
   103  	MOVUPS (A_PTR), X4              \
   104  	MOVUPS 2*SIZE(A_PTR), X5        \
   105  	MOVUPS (A_PTR)(LDA*1), X6       \
   106  	MOVUPS 2*SIZE(A_PTR)(LDA*1), X7 \
   107  	MULPD  X8, X4                   \
   108  	MULPD  X8, X5                   \
   109  	MULPD  X9, X6                   \
   110  	MULPD  X9, X7                   \
   111  	ADDPD  X4, X0                   \
   112  	ADDPD  X5, X1                   \
   113  	ADDPD  X6, X0                   \
   114  	ADDPD  X7, X1                   \
   115  	ADDQ   $4*SIZE, A_PTR
   116  
   117  #define KERNEL_4x1 \
   118  	MOVUPS (A_PTR), X4       \
   119  	MOVUPS 2*SIZE(A_PTR), X5 \
   120  	MULPD  X8, X4            \
   121  	MULPD  X8, X5            \
   122  	ADDPD  X4, X0            \
   123  	ADDPD  X5, X1            \
   124  	ADDQ   $4*SIZE, A_PTR
   125  
   126  #define STORE4 \
   127  	MOVUPS X0, (Y_PTR)       \
   128  	MOVUPS X1, 2*SIZE(Y_PTR)
   129  
   130  #define STORE4_INC \
   131  	MOVLPD X0, (Y_PTR)           \
   132  	MOVHPD X0, (Y_PTR)(INC_Y*1)  \
   133  	MOVLPD X1, (Y_PTR)(INC_Y*2)  \
   134  	MOVHPD X1, (Y_PTR)(INC3_Y*1)
   135  
   136  #define KERNEL_2x4 \
   137  	MOVUPS (A_PTR), X4         \
   138  	MOVUPS (A_PTR)(LDA*1), X5  \
   139  	MOVUPS (A_PTR)(LDA*2), X6  \
   140  	MOVUPS (A_PTR)(LDA3*1), X7 \
   141  	MULPD  X8, X4              \
   142  	MULPD  X9, X5              \
   143  	MULPD  X10, X6             \
   144  	MULPD  X11, X7             \
   145  	ADDPD  X4, X0              \
   146  	ADDPD  X5, X0              \
   147  	ADDPD  X6, X0              \
   148  	ADDPD  X7, X0              \
   149  	ADDQ   $2*SIZE, A_PTR
   150  
   151  #define KERNEL_2x2 \
   152  	MOVUPS (A_PTR), X4        \
   153  	MOVUPS (A_PTR)(LDA*1), X5 \
   154  	MULPD  X8, X4             \
   155  	MULPD  X9, X5             \
   156  	ADDPD  X4, X0             \
   157  	ADDPD  X5, X0             \
   158  	ADDQ   $2*SIZE, A_PTR
   159  
   160  #define KERNEL_2x1 \
   161  	MOVUPS (A_PTR), X4    \
   162  	MULPD  X8, X4         \
   163  	ADDPD  X4, X0         \
   164  	ADDQ   $2*SIZE, A_PTR
   165  
   166  #define STORE2 \
   167  	MOVUPS X0, (Y_PTR)
   168  
   169  #define STORE2_INC \
   170  	MOVLPD X0, (Y_PTR)          \
   171  	MOVHPD X0, (Y_PTR)(INC_Y*1)
   172  
   173  #define KERNEL_1x4 \
   174  	MOVSD (Y_PTR), X0         \
   175  	MOVSD (A_PTR), X4         \
   176  	MOVSD (A_PTR)(LDA*1), X5  \
   177  	MOVSD (A_PTR)(LDA*2), X6  \
   178  	MOVSD (A_PTR)(LDA3*1), X7 \
   179  	MULSD X8, X4              \
   180  	MULSD X9, X5              \
   181  	MULSD X10, X6             \
   182  	MULSD X11, X7             \
   183  	ADDSD X4, X0              \
   184  	ADDSD X5, X0              \
   185  	ADDSD X6, X0              \
   186  	ADDSD X7, X0              \
   187  	MOVSD X0, (Y_PTR)         \
   188  	ADDQ  $SIZE, A_PTR
   189  
   190  #define KERNEL_1x2 \
   191  	MOVSD (Y_PTR), X0        \
   192  	MOVSD (A_PTR), X4        \
   193  	MOVSD (A_PTR)(LDA*1), X5 \
   194  	MULSD X8, X4             \
   195  	MULSD X9, X5             \
   196  	ADDSD X4, X0             \
   197  	ADDSD X5, X0             \
   198  	MOVSD X0, (Y_PTR)        \
   199  	ADDQ  $SIZE, A_PTR
   200  
   201  #define KERNEL_1x1 \
   202  	MOVSD (Y_PTR), X0  \
   203  	MOVSD (A_PTR), X4  \
   204  	MULSD X8, X4       \
   205  	ADDSD X4, X0       \
   206  	MOVSD X0, (Y_PTR)  \
   207  	ADDQ  $SIZE, A_PTR
   208  
   209  #define SCALE_8(PTR, SCAL) \
   210  	MOVUPS (PTR), X0   \
   211  	MOVUPS 16(PTR), X1 \
   212  	MOVUPS 32(PTR), X2 \
   213  	MOVUPS 48(PTR), X3 \
   214  	MULPD  SCAL, X0    \
   215  	MULPD  SCAL, X1    \
   216  	MULPD  SCAL, X2    \
   217  	MULPD  SCAL, X3    \
   218  	MOVUPS X0, (PTR)   \
   219  	MOVUPS X1, 16(PTR) \
   220  	MOVUPS X2, 32(PTR) \
   221  	MOVUPS X3, 48(PTR)
   222  
   223  #define SCALE_4(PTR, SCAL) \
   224  	MOVUPS (PTR), X0   \
   225  	MOVUPS 16(PTR), X1 \
   226  	MULPD  SCAL, X0    \
   227  	MULPD  SCAL, X1    \
   228  	MOVUPS X0, (PTR)   \
   229  	MOVUPS X1, 16(PTR) \
   230  
   231  #define SCALE_2(PTR, SCAL) \
   232  	MOVUPS (PTR), X0 \
   233  	MULPD  SCAL, X0  \
   234  	MOVUPS X0, (PTR) \
   235  
   236  #define SCALE_1(PTR, SCAL) \
   237  	MOVSD (PTR), X0 \
   238  	MULSD SCAL, X0  \
   239  	MOVSD X0, (PTR) \
   240  
   241  #define SCALEINC_4(PTR, INC, INC3, SCAL) \
   242  	MOVSD (PTR), X0         \
   243  	MOVSD (PTR)(INC*1), X1  \
   244  	MOVSD (PTR)(INC*2), X2  \
   245  	MOVSD (PTR)(INC3*1), X3 \
   246  	MULSD SCAL, X0          \
   247  	MULSD SCAL, X1          \
   248  	MULSD SCAL, X2          \
   249  	MULSD SCAL, X3          \
   250  	MOVSD X0, (PTR)         \
   251  	MOVSD X1, (PTR)(INC*1)  \
   252  	MOVSD X2, (PTR)(INC*2)  \
   253  	MOVSD X3, (PTR)(INC3*1)
   254  
   255  #define SCALEINC_2(PTR, INC, SCAL) \
   256  	MOVSD (PTR), X0        \
   257  	MOVSD (PTR)(INC*1), X1 \
   258  	MULSD SCAL, X0         \
   259  	MULSD SCAL, X1         \
   260  	MOVSD X0, (PTR)        \
   261  	MOVSD X1, (PTR)(INC*1)
   262  
   263  // func GemvT(m, n int,
   264  //	alpha float64,
   265  //	a []float64, lda int,
   266  //	x []float64, incX int,
   267  //	beta float64,
   268  //	y []float64, incY int)
   269  TEXT ·GemvT(SB), NOSPLIT, $32-128
   270  	MOVQ M_DIM, M
   271  	MOVQ N_DIM, N
   272  	CMPQ M, $0
   273  	JE   end
   274  	CMPQ N, $0
   275  	JE   end
   276  
   277  	MOVDDUP alpha+16(FP), ALPHA
   278  
   279  	MOVQ x_base+56(FP), X_PTR
   280  	MOVQ y_base+96(FP), Y_PTR
   281  	MOVQ a_base+24(FP), A_ROW
   282  	MOVQ incY+120(FP), INC_Y  // INC_Y = incY * sizeof(float64)
   283  	MOVQ lda+48(FP), LDA      // LDA = LDA * sizeof(float64)
   284  	SHLQ $3, LDA
   285  	LEAQ (LDA)(LDA*2), LDA3   // LDA3 = LDA * 3
   286  	MOVQ A_ROW, A_PTR
   287  
   288  	MOVQ incX+80(FP), INC_X // INC_X = incX * sizeof(float64)
   289  
   290  	XORQ    TMP2, TMP2
   291  	MOVQ    N, TMP1
   292  	SUBQ    $1, TMP1
   293  	NEGQ    TMP1
   294  	IMULQ   INC_X, TMP1
   295  	CMPQ    INC_X, $0
   296  	CMOVQLT TMP1, TMP2
   297  	LEAQ    (X_PTR)(TMP2*SIZE), X_PTR
   298  	MOVQ    X_PTR, X
   299  
   300  	SHLQ $3, INC_X
   301  	LEAQ (INC_X)(INC_X*2), INC3_X // INC3_X = INC_X * 3
   302  
   303  	CMPQ incY+120(FP), $1 // Check for dense vector Y (fast-path)
   304  	JNE  inc
   305  
   306  	MOVSD  $1.0, X0
   307  	COMISD beta+88(FP), X0
   308  	JE     gemv_start
   309  
   310  	MOVSD  $0.0, X0
   311  	COMISD beta+88(FP), X0
   312  	JE     gemv_clear
   313  
   314  	MOVDDUP beta+88(FP), BETA
   315  	SHRQ    $3, M
   316  	JZ      scal4
   317  
   318  scal8:
   319  	SCALE_8(Y_PTR, BETA)
   320  	ADDQ $8*SIZE, Y_PTR
   321  	DECQ M
   322  	JNZ  scal8
   323  
   324  scal4:
   325  	TESTQ $4, M_DIM
   326  	JZ    scal2
   327  	SCALE_4(Y_PTR, BETA)
   328  	ADDQ  $4*SIZE, Y_PTR
   329  
   330  scal2:
   331  	TESTQ $2, M_DIM
   332  	JZ    scal1
   333  	SCALE_2(Y_PTR, BETA)
   334  	ADDQ  $2*SIZE, Y_PTR
   335  
   336  scal1:
   337  	TESTQ $1, M_DIM
   338  	JZ    prep_end
   339  	SCALE_1(Y_PTR, BETA)
   340  
   341  	JMP prep_end
   342  
   343  gemv_clear: // beta == 0 is special cased to clear memory (no nan handling)
   344  	XORPS X0, X0
   345  	XORPS X1, X1
   346  	XORPS X2, X2
   347  	XORPS X3, X3
   348  
   349  	SHRQ $3, M
   350  	JZ   clear4
   351  
   352  clear8:
   353  	MOVUPS X0, (Y_PTR)
   354  	MOVUPS X1, 16(Y_PTR)
   355  	MOVUPS X2, 32(Y_PTR)
   356  	MOVUPS X3, 48(Y_PTR)
   357  	ADDQ   $8*SIZE, Y_PTR
   358  	DECQ   M
   359  	JNZ    clear8
   360  
   361  clear4:
   362  	TESTQ  $4, M_DIM
   363  	JZ     clear2
   364  	MOVUPS X0, (Y_PTR)
   365  	MOVUPS X1, 16(Y_PTR)
   366  	ADDQ   $4*SIZE, Y_PTR
   367  
   368  clear2:
   369  	TESTQ  $2, M_DIM
   370  	JZ     clear1
   371  	MOVUPS X0, (Y_PTR)
   372  	ADDQ   $2*SIZE, Y_PTR
   373  
   374  clear1:
   375  	TESTQ $1, M_DIM
   376  	JZ    prep_end
   377  	MOVSD X0, (Y_PTR)
   378  
   379  prep_end:
   380  	MOVQ Y, Y_PTR
   381  	MOVQ M_DIM, M
   382  
   383  gemv_start:
   384  	SHRQ $2, N
   385  	JZ   c2
   386  
   387  c4:
   388  	// LOAD 4
   389  	INIT4
   390  
   391  	MOVQ M_DIM, M
   392  	SHRQ $2, M
   393  	JZ   c4r2
   394  
   395  c4r4:
   396  	// 4x4 KERNEL
   397  	KERNEL_LOAD4
   398  	KERNEL_4x4
   399  	STORE4
   400  
   401  	ADDQ $4*SIZE, Y_PTR
   402  
   403  	DECQ M
   404  	JNZ  c4r4
   405  
   406  c4r2:
   407  	TESTQ $2, M_DIM
   408  	JZ    c4r1
   409  
   410  	// 4x2 KERNEL
   411  	KERNEL_LOAD2
   412  	KERNEL_2x4
   413  	STORE2
   414  
   415  	ADDQ $2*SIZE, Y_PTR
   416  
   417  c4r1:
   418  	TESTQ $1, M_DIM
   419  	JZ    c4end
   420  
   421  	// 4x1 KERNEL
   422  	KERNEL_1x4
   423  
   424  	ADDQ $SIZE, Y_PTR
   425  
   426  c4end:
   427  	LEAQ (X_PTR)(INC_X*4), X_PTR
   428  	MOVQ Y, Y_PTR
   429  	LEAQ (A_ROW)(LDA*4), A_ROW
   430  	MOVQ A_ROW, A_PTR
   431  
   432  	DECQ N
   433  	JNZ  c4
   434  
   435  c2:
   436  	TESTQ $2, N_DIM
   437  	JZ    c1
   438  
   439  	// LOAD 2
   440  	INIT2
   441  
   442  	MOVQ M_DIM, M
   443  	SHRQ $2, M
   444  	JZ   c2r2
   445  
   446  c2r4:
   447  	// 2x4 KERNEL
   448  	KERNEL_LOAD4
   449  	KERNEL_4x2
   450  	STORE4
   451  
   452  	ADDQ $4*SIZE, Y_PTR
   453  
   454  	DECQ M
   455  	JNZ  c2r4
   456  
   457  c2r2:
   458  	TESTQ $2, M_DIM
   459  	JZ    c2r1
   460  
   461  	// 2x2 KERNEL
   462  	KERNEL_LOAD2
   463  	KERNEL_2x2
   464  	STORE2
   465  
   466  	ADDQ $2*SIZE, Y_PTR
   467  
   468  c2r1:
   469  	TESTQ $1, M_DIM
   470  	JZ    c2end
   471  
   472  	// 2x1 KERNEL
   473  	KERNEL_1x2
   474  
   475  	ADDQ $SIZE, Y_PTR
   476  
   477  c2end:
   478  	LEAQ (X_PTR)(INC_X*2), X_PTR
   479  	MOVQ Y, Y_PTR
   480  	LEAQ (A_ROW)(LDA*2), A_ROW
   481  	MOVQ A_ROW, A_PTR
   482  
   483  c1:
   484  	TESTQ $1, N_DIM
   485  	JZ    end
   486  
   487  	// LOAD 1
   488  	INIT1
   489  
   490  	MOVQ M_DIM, M
   491  	SHRQ $2, M
   492  	JZ   c1r2
   493  
   494  c1r4:
   495  	// 1x4 KERNEL
   496  	KERNEL_LOAD4
   497  	KERNEL_4x1
   498  	STORE4
   499  
   500  	ADDQ $4*SIZE, Y_PTR
   501  
   502  	DECQ M
   503  	JNZ  c1r4
   504  
   505  c1r2:
   506  	TESTQ $2, M_DIM
   507  	JZ    c1r1
   508  
   509  	// 1x2 KERNEL
   510  	KERNEL_LOAD2
   511  	KERNEL_2x1
   512  	STORE2
   513  
   514  	ADDQ $2*SIZE, Y_PTR
   515  
   516  c1r1:
   517  	TESTQ $1, M_DIM
   518  	JZ    end
   519  
   520  	// 1x1 KERNEL
   521  	KERNEL_1x1
   522  
   523  end:
   524  	RET
   525  
   526  inc:  // Algorithm for incX != 0 ( split loads in kernel )
   527  	XORQ    TMP2, TMP2
   528  	MOVQ    M, TMP1
   529  	SUBQ    $1, TMP1
   530  	IMULQ   INC_Y, TMP1
   531  	NEGQ    TMP1
   532  	CMPQ    INC_Y, $0
   533  	CMOVQLT TMP1, TMP2
   534  	LEAQ    (Y_PTR)(TMP2*SIZE), Y_PTR
   535  	MOVQ    Y_PTR, Y
   536  
   537  	SHLQ $3, INC_Y
   538  	LEAQ (INC_Y)(INC_Y*2), INC3_Y // INC3_Y = INC_Y * 3
   539  
   540  	MOVSD  $1.0, X0
   541  	COMISD beta+88(FP), X0
   542  	JE     inc_gemv_start
   543  
   544  	MOVSD  $0.0, X0
   545  	COMISD beta+88(FP), X0
   546  	JE     inc_gemv_clear
   547  
   548  	MOVDDUP beta+88(FP), BETA
   549  	SHRQ    $2, M
   550  	JZ      inc_scal2
   551  
   552  inc_scal4:
   553  	SCALEINC_4(Y_PTR, INC_Y, INC3_Y, BETA)
   554  	LEAQ (Y_PTR)(INC_Y*4), Y_PTR
   555  	DECQ M
   556  	JNZ  inc_scal4
   557  
   558  inc_scal2:
   559  	TESTQ $2, M_DIM
   560  	JZ    inc_scal1
   561  
   562  	SCALEINC_2(Y_PTR, INC_Y, BETA)
   563  	LEAQ (Y_PTR)(INC_Y*2), Y_PTR
   564  
   565  inc_scal1:
   566  	TESTQ $1, M_DIM
   567  	JZ    inc_prep_end
   568  	SCALE_1(Y_PTR, BETA)
   569  
   570  	JMP inc_prep_end
   571  
   572  inc_gemv_clear: // beta == 0 is special-cased to clear memory (no nan handling)
   573  	XORPS X0, X0
   574  	XORPS X1, X1
   575  	XORPS X2, X2
   576  	XORPS X3, X3
   577  
   578  	SHRQ $2, M
   579  	JZ   inc_clear2
   580  
   581  inc_clear4:
   582  	MOVSD X0, (Y_PTR)
   583  	MOVSD X1, (Y_PTR)(INC_Y*1)
   584  	MOVSD X2, (Y_PTR)(INC_Y*2)
   585  	MOVSD X3, (Y_PTR)(INC3_Y*1)
   586  	LEAQ  (Y_PTR)(INC_Y*4), Y_PTR
   587  	DECQ  M
   588  	JNZ   inc_clear4
   589  
   590  inc_clear2:
   591  	TESTQ $2, M_DIM
   592  	JZ    inc_clear1
   593  	MOVSD X0, (Y_PTR)
   594  	MOVSD X1, (Y_PTR)(INC_Y*1)
   595  	LEAQ  (Y_PTR)(INC_Y*2), Y_PTR
   596  
   597  inc_clear1:
   598  	TESTQ $1, M_DIM
   599  	JZ    inc_prep_end
   600  	MOVSD X0, (Y_PTR)
   601  
   602  inc_prep_end:
   603  	MOVQ Y, Y_PTR
   604  	MOVQ M_DIM, M
   605  
   606  inc_gemv_start:
   607  	SHRQ $2, N
   608  	JZ   inc_c2
   609  
   610  inc_c4:
   611  	// LOAD 4
   612  	INIT4
   613  
   614  	MOVQ M_DIM, M
   615  	SHRQ $2, M
   616  	JZ   inc_c4r2
   617  
   618  inc_c4r4:
   619  	// 4x4 KERNEL
   620  	KERNEL_LOAD4_INC
   621  	KERNEL_4x4
   622  	STORE4_INC
   623  
   624  	LEAQ (Y_PTR)(INC_Y*4), Y_PTR
   625  
   626  	DECQ M
   627  	JNZ  inc_c4r4
   628  
   629  inc_c4r2:
   630  	TESTQ $2, M_DIM
   631  	JZ    inc_c4r1
   632  
   633  	// 4x2 KERNEL
   634  	KERNEL_LOAD2_INC
   635  	KERNEL_2x4
   636  	STORE2_INC
   637  
   638  	LEAQ (Y_PTR)(INC_Y*2), Y_PTR
   639  
   640  inc_c4r1:
   641  	TESTQ $1, M_DIM
   642  	JZ    inc_c4end
   643  
   644  	// 4x1 KERNEL
   645  	KERNEL_1x4
   646  
   647  	ADDQ INC_Y, Y_PTR
   648  
   649  inc_c4end:
   650  	LEAQ (X_PTR)(INC_X*4), X_PTR
   651  	MOVQ Y, Y_PTR
   652  	LEAQ (A_ROW)(LDA*4), A_ROW
   653  	MOVQ A_ROW, A_PTR
   654  
   655  	DECQ N
   656  	JNZ  inc_c4
   657  
   658  inc_c2:
   659  	TESTQ $2, N_DIM
   660  	JZ    inc_c1
   661  
   662  	// LOAD 2
   663  	INIT2
   664  
   665  	MOVQ M_DIM, M
   666  	SHRQ $2, M
   667  	JZ   inc_c2r2
   668  
   669  inc_c2r4:
   670  	// 2x4 KERNEL
   671  	KERNEL_LOAD4_INC
   672  	KERNEL_4x2
   673  	STORE4_INC
   674  
   675  	LEAQ (Y_PTR)(INC_Y*4), Y_PTR
   676  	DECQ M
   677  	JNZ  inc_c2r4
   678  
   679  inc_c2r2:
   680  	TESTQ $2, M_DIM
   681  	JZ    inc_c2r1
   682  
   683  	// 2x2 KERNEL
   684  	KERNEL_LOAD2_INC
   685  	KERNEL_2x2
   686  	STORE2_INC
   687  
   688  	LEAQ (Y_PTR)(INC_Y*2), Y_PTR
   689  
   690  inc_c2r1:
   691  	TESTQ $1, M_DIM
   692  	JZ    inc_c2end
   693  
   694  	// 2x1 KERNEL
   695  	KERNEL_1x2
   696  
   697  	ADDQ INC_Y, Y_PTR
   698  
   699  inc_c2end:
   700  	LEAQ (X_PTR)(INC_X*2), X_PTR
   701  	MOVQ Y, Y_PTR
   702  	LEAQ (A_ROW)(LDA*2), A_ROW
   703  	MOVQ A_ROW, A_PTR
   704  
   705  inc_c1:
   706  	TESTQ $1, N_DIM
   707  	JZ    inc_end
   708  
   709  	// LOAD 1
   710  	INIT1
   711  
   712  	MOVQ M_DIM, M
   713  	SHRQ $2, M
   714  	JZ   inc_c1r2
   715  
   716  inc_c1r4:
   717  	// 1x4 KERNEL
   718  	KERNEL_LOAD4_INC
   719  	KERNEL_4x1
   720  	STORE4_INC
   721  
   722  	LEAQ (Y_PTR)(INC_Y*4), Y_PTR
   723  	DECQ M
   724  	JNZ  inc_c1r4
   725  
   726  inc_c1r2:
   727  	TESTQ $2, M_DIM
   728  	JZ    inc_c1r1
   729  
   730  	// 1x2 KERNEL
   731  	KERNEL_LOAD2_INC
   732  	KERNEL_2x1
   733  	STORE2_INC
   734  
   735  	LEAQ (Y_PTR)(INC_Y*2), Y_PTR
   736  
   737  inc_c1r1:
   738  	TESTQ $1, M_DIM
   739  	JZ    inc_end
   740  
   741  	// 1x1 KERNEL
   742  	KERNEL_1x1
   743  
   744  inc_end:
   745  	RET