libflame revision_anchor
Functions
FLA_Apply_CAQ2_UT_lhfc.h File Reference

(r)

Go to the source code of this file.

Functions

FLA_Error FLA_Apply_CAQ2_UT_lhfc_blk_var1 (FLA_Obj D, FLA_Obj T, FLA_Obj W1, FLA_Obj C, FLA_Obj E, fla_apcaq2ut_t *cntl)
 
FLA_Error FLA_Apply_CAQ2_UT_lhfc_blk_var2 (FLA_Obj D, FLA_Obj T, FLA_Obj W1, FLA_Obj C, FLA_Obj E, fla_apcaq2ut_t *cntl)
 
FLA_Error FLA_Apply_CAQ2_UT_lhfc_blk_var3 (FLA_Obj D, FLA_Obj T, FLA_Obj W, FLA_Obj C, FLA_Obj E, fla_apcaq2ut_t *cntl)
 

Function Documentation

◆ FLA_Apply_CAQ2_UT_lhfc_blk_var1()

FLA_Error FLA_Apply_CAQ2_UT_lhfc_blk_var1 ( FLA_Obj  D,
FLA_Obj  T,
FLA_Obj  W1,
FLA_Obj  C,
FLA_Obj  E,
fla_apcaq2ut_t cntl 
)
15{
16 FLA_Obj DTL, DTR, D00, D01, D02,
17 DBL, DBR, D10, D11, D12,
18 D20, D21, D22;
19
20 FLA_Obj TL, TR, T0, T1, T2;
21
23 T2B;
24
25 FLA_Obj CT, C0,
26 CB, C1,
27 C2;
28
29 FLA_Obj ET, E0,
30 EB, E1,
31 E2;
32
34 W1BL, W1BR;
35
36 dim_t b_alg, b;
37
38 // Query the algorithmic blocksize by inspecting the length of T.
40
41 FLA_Part_2x2( D, &DTL, &DTR,
42 &DBL, &DBR, 0, 0, FLA_TL );
43
44 FLA_Part_1x2( T, &TL, &TR, 0, FLA_LEFT );
45
46 FLA_Part_2x1( C, &CT,
47 &CB, 0, FLA_TOP );
48
49 FLA_Part_2x1( E, &ET,
50 &EB, 0, FLA_TOP );
51
52 while ( FLA_Obj_width( DBR ) > 0 ){
53
54 b = min( b_alg, FLA_Obj_width( DBR ) );
55
56 FLA_Repart_2x2_to_3x3( DTL, /**/ DTR, &D00, /**/ &D01, &D02,
57 /* ************* */ /* ******************** */
58 &D10, /**/ &D11, &D12,
59 DBL, /**/ DBR, &D20, /**/ &D21, &D22,
60 b, b, FLA_BR );
61
62 FLA_Repart_1x2_to_1x3( TL, /**/ TR, &T0, /**/ &T1, &T2,
63 b, FLA_RIGHT );
64
66 /* ** */ /* ** */
67 &C1,
68 CB, &C2, b, FLA_BOTTOM );
69
71 /* ** */ /* ** */
72 &E1,
73 EB, &E2, b, FLA_BOTTOM );
74
75 /*------------------------------------------------------------*/
76
78 &T2B, b, FLA_TOP );
79
81 &W1BL, &W1BR, b, FLA_Obj_width( C1 ), FLA_TL );
82
83 // W1TL = inv( triu( T1T ) )' * ( C1 + D1' * E );
84 // = inv( triu( T1T ) )' * ( C1 + D01' * E0 + D11' * E1 );
85
87 FLA_Cntl_sub_copy( cntl ) );
88
92 FLA_Cntl_sub_trmm1( cntl ) );
93
96 FLA_Cntl_sub_gemm1( cntl ) );
97
99 FLA_Cntl_sub_axpy1( cntl ) );
100
103 FLA_ONE, T1T, W1TL,
104 FLA_Cntl_sub_trsm( cntl ) );
105
106 // C1 = C1 - W1TL;
107 // E = E - D1 * W1TL;
108 // => E0 = E0 - D01 * W1TL;
109 // E1 = E1 - D11 * W1TL;
110
112 FLA_Cntl_sub_axpy2( cntl ) );
113
116 FLA_Cntl_sub_gemm2( cntl ) );
117
120 FLA_ONE, D11, W1TL,
121 FLA_Cntl_sub_trmm2( cntl ) );
122
124 FLA_Cntl_sub_axpy3( cntl ) );
125
126 /*------------------------------------------------------------*/
127
128 FLA_Cont_with_3x3_to_2x2( &DTL, /**/ &DTR, D00, D01, /**/ D02,
129 D10, D11, /**/ D12,
130 /* ************** */ /* ****************** */
131 &DBL, /**/ &DBR, D20, D21, /**/ D22,
132 FLA_TL );
133
134 FLA_Cont_with_1x3_to_1x2( &TL, /**/ &TR, T0, T1, /**/ T2,
135 FLA_LEFT );
136
138 C1,
139 /* ** */ /* ** */
140 &CB, C2, FLA_TOP );
141
143 E1,
144 /* ** */ /* ** */
145 &EB, E2, FLA_TOP );
146 }
147
148 return FLA_SUCCESS;
149}
FLA_Error FLA_Axpy_internal(FLA_Obj alpha, FLA_Obj A, FLA_Obj B, fla_axpy_t *cntl)
Definition FLA_Axpy_internal.c:16
FLA_Error FLA_Copy_internal(FLA_Obj A, FLA_Obj B, fla_copy_t *cntl)
Definition FLA_Copy_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
FLA_Error FLA_Trmm_internal(FLA_Side side, FLA_Uplo uplo, FLA_Trans transa, FLA_Diag diag, FLA_Obj alpha, FLA_Obj A, FLA_Obj B, fla_trmm_t *cntl)
Definition FLA_Trmm_internal.c:16
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_Obj FLA_MINUS_ONE
Definition FLA_Init.c:22
FLA_Obj FLA_ONE
Definition FLA_Init.c:18
FLA_Error FLA_Cont_with_3x3_to_2x2(FLA_Obj *ATL, FLA_Obj *ATR, FLA_Obj A00, FLA_Obj A01, FLA_Obj A02, FLA_Obj A10, FLA_Obj A11, FLA_Obj A12, FLA_Obj *ABL, FLA_Obj *ABR, FLA_Obj A20, FLA_Obj A21, FLA_Obj A22, FLA_Quadrant quadrant)
Definition FLA_View.c:304
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_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_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_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
dim_t FLA_Obj_width(FLA_Obj obj)
Definition FLA_Query.c:123
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_Part_2x1(FLA_Obj A, FLA_Obj *A1, FLA_Obj *A2, dim_t mb, FLA_Side side)
Definition FLA_View.c:76
dim_t FLA_Obj_length(FLA_Obj obj)
Definition FLA_Query.c:116
FLA_Error FLA_Repart_2x2_to_3x3(FLA_Obj ATL, FLA_Obj ATR, FLA_Obj *A00, FLA_Obj *A01, FLA_Obj *A02, FLA_Obj *A10, FLA_Obj *A11, FLA_Obj *A12, FLA_Obj ABL, FLA_Obj ABR, FLA_Obj *A20, FLA_Obj *A21, FLA_Obj *A22, dim_t mb, dim_t nb, FLA_Quadrant quadrant)
Definition FLA_View.c:142
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
int i
Definition bl1_axmyv2.c:145
Definition FLA_type_defs.h:159

References FLA_Axpy_internal(), FLA_Cont_with_1x3_to_1x2(), FLA_Cont_with_3x1_to_2x1(), FLA_Cont_with_3x3_to_2x2(), FLA_Copy_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(), FLA_Repart_2x2_to_3x3(), FLA_Trmm_internal(), FLA_Trsm_internal(), and i.

Referenced by FLA_Apply_CAQ2_UT_lhfc().

◆ FLA_Apply_CAQ2_UT_lhfc_blk_var2()

FLA_Error FLA_Apply_CAQ2_UT_lhfc_blk_var2 ( FLA_Obj  D,
FLA_Obj  T,
FLA_Obj  W1,
FLA_Obj  C,
FLA_Obj  E,
fla_apcaq2ut_t cntl 
)
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
42
44 /* ** */ /* ** */
45 &D1,
46 DB, &D2, b, FLA_BOTTOM );
47
49 /* ** */ /* ** */
50 &T1,
51 TB, &T2, b, FLA_BOTTOM );
52
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
66 D1, T1, W1, C,
67 E1, FLA_Cntl_sub_apcaq2ut( cntl ) );
68
69 /*------------------------------------------------------------*/
70
72 D1,
73 /* ** */ /* ** */
74 &DB, D2, FLA_TOP );
75
77 T1,
78 /* ** */ /* ** */
79 &TB, T2, FLA_TOP );
80
82 E1,
83 /* ** */ /* ** */
84 &EB, E2, FLA_TOP );
85 }
86
87 return FLA_SUCCESS;
88}
FLA_Error FLA_Apply_CAQ2_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_apcaq2ut_t *cntl)
Definition FLA_Apply_CAQ2_UT_internal.c:17
dim_t FLA_Determine_blocksize(FLA_Obj A_unproc, FLA_Quadrant to_dir, fla_blocksize_t *cntl_blocksizes)
Definition FLA_Blocksize.c:234

References FLA_Apply_CAQ2_UT_internal(), FLA_Cont_with_3x1_to_2x1(), FLA_Determine_blocksize(), FLA_Obj_length(), FLA_Part_2x1(), FLA_Repart_2x1_to_3x1(), and i.

Referenced by FLA_Apply_CAQ2_UT_lhfc().

◆ FLA_Apply_CAQ2_UT_lhfc_blk_var3()

FLA_Error FLA_Apply_CAQ2_UT_lhfc_blk_var3 ( FLA_Obj  D,
FLA_Obj  T,
FLA_Obj  W,
FLA_Obj  C,
FLA_Obj  E,
fla_apcaq2ut_t cntl 
)
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
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
51 D, T, W1, C1,
52 E1, FLA_Cntl_sub_apcaq2ut( 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}

References FLA_Apply_CAQ2_UT_internal(), FLA_Cont_with_1x3_to_1x2(), FLA_Determine_blocksize(), FLA_Obj_width(), FLA_Part_1x2(), FLA_Repart_1x2_to_1x3(), and i.

Referenced by FLA_Apply_CAQ2_UT_lhfc().