libflame  revision_anchor
Functions
FLA_Apply_Q2_UT_lhfc.h File Reference

(r)

Go to the source code of this file.

Functions

FLA_Error FLA_Apply_Q2_UT_lhfc_blk_var1 (FLA_Obj D, FLA_Obj T, FLA_Obj W1, FLA_Obj C, FLA_Obj E, fla_apq2ut_t *cntl)
 
FLA_Error FLA_Apply_Q2_UT_lhfc_blk_var2 (FLA_Obj D, FLA_Obj T, FLA_Obj W1, FLA_Obj C, FLA_Obj E, fla_apq2ut_t *cntl)
 
FLA_Error FLA_Apply_Q2_UT_lhfc_blk_var3 (FLA_Obj D, FLA_Obj T, FLA_Obj W, FLA_Obj C, FLA_Obj E, fla_apq2ut_t *cntl)
 

Function Documentation

◆ FLA_Apply_Q2_UT_lhfc_blk_var1()

FLA_Error FLA_Apply_Q2_UT_lhfc_blk_var1 ( FLA_Obj  D,
FLA_Obj  T,
FLA_Obj  W1,
FLA_Obj  C,
FLA_Obj  E,
fla_apq2ut_t cntl 
)

References FLA_Axpyt_internal(), FLA_Cont_with_1x3_to_1x2(), FLA_Cont_with_3x1_to_2x1(), FLA_Copyt_internal(), FLA_Gemm_internal(), FLA_MINUS_ONE, FLA_Obj_length(), FLA_Obj_width(), FLA_ONE, FLA_Part_1x2(), FLA_Part_2x1(), FLA_Part_2x2(), FLA_Repart_1x2_to_1x3(), FLA_Repart_2x1_to_3x1(), and FLA_Trsm_internal().

Referenced by FLA_Apply_Q2_UT_lhfc().

15 {
16  FLA_Obj DL, DR, D0, D1, D2;
17 
18  FLA_Obj TL, TR, T0, T1, T2;
19 
20  FLA_Obj T1T,
21  T2B;
22 
23  FLA_Obj CT, C0,
24  CB, C1,
25  C2;
26 
27  FLA_Obj W1TL, W1TR,
28  W1BL, W1BR;
29 
30  dim_t b_alg, b;
31 
32  // Query the algorithmic blocksize by inspecting the length of T.
33  b_alg = FLA_Obj_length( T );
34 
35  FLA_Part_1x2( D, &DL, &DR, 0, FLA_LEFT );
36 
37  FLA_Part_1x2( T, &TL, &TR, 0, FLA_LEFT );
38 
39  FLA_Part_2x1( C, &CT,
40  &CB, 0, FLA_TOP );
41 
42  while ( FLA_Obj_width( DL ) < FLA_Obj_width( D ) ){
43 
44  b = min( b_alg, FLA_Obj_width( DR ) );
45 
46  FLA_Repart_1x2_to_1x3( DL, /**/ DR, &D0, /**/ &D1, &D2,
47  b, FLA_RIGHT );
48 
49  FLA_Repart_1x2_to_1x3( TL, /**/ TR, &T0, /**/ &T1, &T2,
50  b, FLA_RIGHT );
51 
52  FLA_Repart_2x1_to_3x1( CT, &C0,
53  /* ** */ /* ** */
54  &C1,
55  CB, &C2, b, FLA_BOTTOM );
56 
57  /*------------------------------------------------------------*/
58 
59  FLA_Part_2x1( T1, &T1T,
60  &T2B, b, FLA_TOP );
61 
62  FLA_Part_2x2( W1, &W1TL, &W1TR,
63  &W1BL, &W1BR, b, FLA_Obj_width( C1 ), FLA_TL );
64 
65  // W1TL = C1;
66 
67  FLA_Copyt_internal( FLA_NO_TRANSPOSE, C1, W1TL,
68  FLA_Cntl_sub_copyt( cntl ) );
69 
70  // W1TL = inv( triu( T1T ) )' * ( C1 + D1' * E );
71 
72  FLA_Gemm_internal( FLA_CONJ_TRANSPOSE, FLA_NO_TRANSPOSE,
73  FLA_ONE, D1, E, FLA_ONE, W1TL,
74  FLA_Cntl_sub_gemm1( cntl ) );
75 
76  FLA_Trsm_internal( FLA_LEFT, FLA_UPPER_TRIANGULAR,
77  FLA_CONJ_TRANSPOSE, FLA_NONUNIT_DIAG,
78  FLA_ONE, T1T, W1TL,
79  FLA_Cntl_sub_trsm( cntl ) );
80 
81  // C1 = C1 - W1TL;
82  // E = E - D1 * W1TL;
83 
84  FLA_Axpyt_internal( FLA_NO_TRANSPOSE, FLA_MINUS_ONE, W1TL, C1,
85  FLA_Cntl_sub_axpyt( cntl ) );
86 
87  FLA_Gemm_internal( FLA_NO_TRANSPOSE, FLA_NO_TRANSPOSE,
88  FLA_MINUS_ONE, D1, W1TL, FLA_ONE, E,
89  FLA_Cntl_sub_gemm2( cntl ) );
90 
91  /*------------------------------------------------------------*/
92 
93  FLA_Cont_with_1x3_to_1x2( &DL, /**/ &DR, D0, D1, /**/ D2,
94  FLA_LEFT );
95 
96  FLA_Cont_with_1x3_to_1x2( &TL, /**/ &TR, T0, T1, /**/ T2,
97  FLA_LEFT );
98 
99  FLA_Cont_with_3x1_to_2x1( &CT, C0,
100  C1,
101  /* ** */ /* ** */
102  &CB, C2, FLA_TOP );
103  }
104 
105  return FLA_SUCCESS;
106 }
FLA_Error FLA_Repart_2x1_to_3x1(FLA_Obj AT, FLA_Obj *A0, FLA_Obj *A1, FLA_Obj AB, FLA_Obj *A2, dim_t mb, FLA_Side side)
Definition: FLA_View.c:226
FLA_Error FLA_Repart_1x2_to_1x3(FLA_Obj AL, FLA_Obj AR, FLA_Obj *A0, FLA_Obj *A1, FLA_Obj *A2, dim_t nb, FLA_Side side)
Definition: FLA_View.c:267
FLA_Error FLA_Copyt_internal(FLA_Trans trans, FLA_Obj A, FLA_Obj B, fla_copyt_t *cntl)
Definition: FLA_Copyt_internal.c:16
FLA_Error FLA_Gemm_internal(FLA_Trans transa, FLA_Trans transb, FLA_Obj alpha, FLA_Obj A, FLA_Obj B, FLA_Obj beta, FLA_Obj C, fla_gemm_t *cntl)
Definition: FLA_Gemm_internal.c:16
unsigned long dim_t
Definition: FLA_type_defs.h:71
FLA_Error FLA_Axpyt_internal(FLA_Trans trans, FLA_Obj alpha, FLA_Obj A, FLA_Obj B, fla_axpyt_t *cntl)
Definition: FLA_Axpyt_internal.c:16
FLA_Obj FLA_MINUS_ONE
Definition: FLA_Init.c:22
FLA_Error FLA_Cont_with_3x1_to_2x1(FLA_Obj *AT, FLA_Obj A0, FLA_Obj A1, FLA_Obj *AB, FLA_Obj A2, FLA_Side side)
Definition: FLA_View.c:428
FLA_Error FLA_Part_2x2(FLA_Obj A, FLA_Obj *A11, FLA_Obj *A12, FLA_Obj *A21, FLA_Obj *A22, dim_t mb, dim_t nb, FLA_Quadrant quadrant)
Definition: FLA_View.c:17
FLA_Obj FLA_ONE
Definition: FLA_Init.c:18
Definition: FLA_type_defs.h:158
dim_t FLA_Obj_width(FLA_Obj obj)
Definition: FLA_Query.c:123
FLA_Error FLA_Cont_with_1x3_to_1x2(FLA_Obj *AL, FLA_Obj *AR, FLA_Obj A0, FLA_Obj A1, FLA_Obj A2, FLA_Side side)
Definition: FLA_View.c:475
FLA_Error FLA_Part_2x1(FLA_Obj A, FLA_Obj *A1, FLA_Obj *A2, dim_t mb, FLA_Side side)
Definition: FLA_View.c:76
FLA_Error FLA_Part_1x2(FLA_Obj A, FLA_Obj *A1, FLA_Obj *A2, dim_t nb, FLA_Side side)
Definition: FLA_View.c:110
dim_t FLA_Obj_length(FLA_Obj obj)
Definition: FLA_Query.c:116
FLA_Error FLA_Trsm_internal(FLA_Side side, FLA_Uplo uplo, FLA_Trans transa, FLA_Diag diag, FLA_Obj alpha, FLA_Obj A, FLA_Obj B, fla_trsm_t *cntl)
Definition: FLA_Trsm_internal.c:16

◆ FLA_Apply_Q2_UT_lhfc_blk_var2()

FLA_Error FLA_Apply_Q2_UT_lhfc_blk_var2 ( FLA_Obj  D,
FLA_Obj  T,
FLA_Obj  W1,
FLA_Obj  C,
FLA_Obj  E,
fla_apq2ut_t cntl 
)

References FLA_Apply_Q2_UT_internal(), FLA_Cont_with_3x1_to_2x1(), FLA_Determine_blocksize(), FLA_Obj_length(), FLA_Part_2x1(), and FLA_Repart_2x1_to_3x1().

Referenced by FLA_Apply_Q2_UT_lhfc().

15 {
16  FLA_Obj DT, D0,
17  DB, D1,
18  D2;
19 
20  FLA_Obj TT, T0,
21  TB, T1,
22  T2;
23 
24  FLA_Obj ET, E0,
25  EB, E1,
26  E2;
27 
28  dim_t b;
29 
30  FLA_Part_2x1( D, &DT,
31  &DB, 0, FLA_TOP );
32 
33  FLA_Part_2x1( T, &TT,
34  &TB, 0, FLA_TOP );
35 
36  FLA_Part_2x1( E, &ET,
37  &EB, 0, FLA_TOP );
38 
39  while ( FLA_Obj_length( DT ) < FLA_Obj_length( D ) ){
40 
41  b = FLA_Determine_blocksize( DB, FLA_BOTTOM, FLA_Cntl_blocksize( cntl ) );
42 
43  FLA_Repart_2x1_to_3x1( DT, &D0,
44  /* ** */ /* ** */
45  &D1,
46  DB, &D2, b, FLA_BOTTOM );
47 
48  FLA_Repart_2x1_to_3x1( TT, &T0,
49  /* ** */ /* ** */
50  &T1,
51  TB, &T2, b, FLA_BOTTOM );
52 
53  FLA_Repart_2x1_to_3x1( ET, &E0,
54  /* ** */ /* ** */
55  &E1,
56  EB, &E2, b, FLA_BOTTOM );
57 
58  /*------------------------------------------------------------*/
59 
60  // / C \ = Q' / C \
61  // \ E1 / \ E1 /
62  //
63  // where Q is formed from D1 and T1.
64 
65  FLA_Apply_Q2_UT_internal( FLA_LEFT, FLA_CONJ_TRANSPOSE, FLA_FORWARD, FLA_COLUMNWISE,
66  D1, T1, W1, C,
67  E1, FLA_Cntl_sub_apq2ut( cntl ) );
68 
69  /*------------------------------------------------------------*/
70 
71  FLA_Cont_with_3x1_to_2x1( &DT, D0,
72  D1,
73  /* ** */ /* ** */
74  &DB, D2, FLA_TOP );
75 
76  FLA_Cont_with_3x1_to_2x1( &TT, T0,
77  T1,
78  /* ** */ /* ** */
79  &TB, T2, FLA_TOP );
80 
81  FLA_Cont_with_3x1_to_2x1( &ET, E0,
82  E1,
83  /* ** */ /* ** */
84  &EB, E2, FLA_TOP );
85  }
86 
87  return FLA_SUCCESS;
88 }
FLA_Error FLA_Repart_2x1_to_3x1(FLA_Obj AT, FLA_Obj *A0, FLA_Obj *A1, FLA_Obj AB, FLA_Obj *A2, dim_t mb, FLA_Side side)
Definition: FLA_View.c:226
unsigned long dim_t
Definition: FLA_type_defs.h:71
dim_t FLA_Determine_blocksize(FLA_Obj A_unproc, FLA_Quadrant to_dir, fla_blocksize_t *cntl_blocksizes)
Definition: FLA_Blocksize.c:234
FLA_Error FLA_Cont_with_3x1_to_2x1(FLA_Obj *AT, FLA_Obj A0, FLA_Obj A1, FLA_Obj *AB, FLA_Obj A2, FLA_Side side)
Definition: FLA_View.c:428
Definition: FLA_type_defs.h:158
FLA_Error FLA_Part_2x1(FLA_Obj A, FLA_Obj *A1, FLA_Obj *A2, dim_t mb, FLA_Side side)
Definition: FLA_View.c:76
FLA_Error FLA_Apply_Q2_UT_internal(FLA_Side side, FLA_Trans trans, FLA_Direct direct, FLA_Store storev, FLA_Obj D, FLA_Obj T, FLA_Obj W, FLA_Obj C, FLA_Obj E, fla_apq2ut_t *cntl)
Definition: FLA_Apply_Q2_UT_internal.c:17
dim_t FLA_Obj_length(FLA_Obj obj)
Definition: FLA_Query.c:116

◆ FLA_Apply_Q2_UT_lhfc_blk_var3()

FLA_Error FLA_Apply_Q2_UT_lhfc_blk_var3 ( FLA_Obj  D,
FLA_Obj  T,
FLA_Obj  W,
FLA_Obj  C,
FLA_Obj  E,
fla_apq2ut_t cntl 
)

References FLA_Apply_Q2_UT_internal(), FLA_Cont_with_1x3_to_1x2(), FLA_Determine_blocksize(), FLA_Obj_width(), FLA_Part_1x2(), and FLA_Repart_1x2_to_1x3().

Referenced by FLA_Apply_Q2_UT_lhfc().

15 {
16  FLA_Obj WL, WR, W0, W1, W2;
17 
18  FLA_Obj CL, CR, C0, C1, C2;
19 
20  FLA_Obj EL, ER, E0, E1, E2;
21 
22  dim_t b;
23 
24  FLA_Part_1x2( W, &WL, &WR, 0, FLA_LEFT );
25 
26  FLA_Part_1x2( C, &CL, &CR, 0, FLA_LEFT );
27 
28  FLA_Part_1x2( E, &EL, &ER, 0, FLA_LEFT );
29 
30  while ( FLA_Obj_width( CL ) < FLA_Obj_width( C ) ){
31 
32  b = FLA_Determine_blocksize( CR, FLA_RIGHT, FLA_Cntl_blocksize( cntl ) );
33 
34  FLA_Repart_1x2_to_1x3( WL, /**/ WR, &W0, /**/ &W1, &W2,
35  b, FLA_RIGHT );
36 
37  FLA_Repart_1x2_to_1x3( CL, /**/ CR, &C0, /**/ &C1, &C2,
38  b, FLA_RIGHT );
39 
40  FLA_Repart_1x2_to_1x3( EL, /**/ ER, &E0, /**/ &E1, &E2,
41  b, FLA_RIGHT );
42 
43  /*------------------------------------------------------------*/
44 
45  // / C1 \ = Q' / C1 \
46  // \ E1 / \ E1 /
47  //
48  // where Q is formed from D and T.
49 
50  FLA_Apply_Q2_UT_internal( FLA_LEFT, FLA_CONJ_TRANSPOSE, FLA_FORWARD, FLA_COLUMNWISE,
51  D, T, W1, C1,
52  E1, FLA_Cntl_sub_apq2ut( cntl ) );
53 
54  /*------------------------------------------------------------*/
55 
56  FLA_Cont_with_1x3_to_1x2( &WL, /**/ &WR, W0, W1, /**/ W2,
57  FLA_LEFT );
58 
59  FLA_Cont_with_1x3_to_1x2( &CL, /**/ &CR, C0, C1, /**/ C2,
60  FLA_LEFT );
61 
62  FLA_Cont_with_1x3_to_1x2( &EL, /**/ &ER, E0, E1, /**/ E2,
63  FLA_LEFT );
64  }
65 
66  return FLA_SUCCESS;
67 }
FLA_Error FLA_Repart_1x2_to_1x3(FLA_Obj AL, FLA_Obj AR, FLA_Obj *A0, FLA_Obj *A1, FLA_Obj *A2, dim_t nb, FLA_Side side)
Definition: FLA_View.c:267
unsigned long dim_t
Definition: FLA_type_defs.h:71
dim_t FLA_Determine_blocksize(FLA_Obj A_unproc, FLA_Quadrant to_dir, fla_blocksize_t *cntl_blocksizes)
Definition: FLA_Blocksize.c:234
Definition: FLA_type_defs.h:158
dim_t FLA_Obj_width(FLA_Obj obj)
Definition: FLA_Query.c:123
FLA_Error FLA_Cont_with_1x3_to_1x2(FLA_Obj *AL, FLA_Obj *AR, FLA_Obj A0, FLA_Obj A1, FLA_Obj A2, FLA_Side side)
Definition: FLA_View.c:475
FLA_Error FLA_Part_1x2(FLA_Obj A, FLA_Obj *A1, FLA_Obj *A2, dim_t nb, FLA_Side side)
Definition: FLA_View.c:110
FLA_Error FLA_Apply_Q2_UT_internal(FLA_Side side, FLA_Trans trans, FLA_Direct direct, FLA_Store storev, FLA_Obj D, FLA_Obj T, FLA_Obj W, FLA_Obj C, FLA_Obj E, fla_apq2ut_t *cntl)
Definition: FLA_Apply_Q2_UT_internal.c:17