GCC Code Coverage Report

Directory: ./
File: src/svd/impls/cyclic/cycliccuda/cycliccuda.cu
Date: 2025-10-03 04:28:47
Exec Total Coverage
Lines: 58 58 100.0%
Functions: 2 2 100.0%
Branches: 46 88 52.3%
Line Branch Exec Source
1 3553 /*
2 - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
3 SLEPc - Scalable Library for Eigenvalue Problem Computations
4 Copyright (c) 2002-, Universitat Politecnica de Valencia, Spain
5
6 This file is part of SLEPc.
7 SLEPc is distributed under a 2-clause BSD license (see LICENSE).
8 - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
9 */
10 /*
11 SLEPc singular value solver: "cyclic" (CUDA implementation)
12 */
13 #include <slepc/private/svdimpl.h>
14 #include "../src/svd/impls/cyclic/cyclic.h"
15
16 348 PetscErrorCode MatMult_Cyclic_CUDA(Mat B,Vec x,Vec y)
17 {
18 348 SVD_CYCLIC_SHELL *ctx;
19 348 const PetscScalar *d_px;
20 348 PetscScalar *d_py;
21 348 PetscInt m;
22
23 348 PetscFunctionBegin;
24
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 348 PetscCall(MatShellGetContext(B,&ctx));
25
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 348 PetscCall(MatGetLocalSize(ctx->A,&m,NULL));
26
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 348 PetscCall(VecCUDAGetArrayRead(x,&d_px));
27
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 348 PetscCall(VecCUDAGetArrayWrite(y,&d_py));
28
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 348 PetscCall(VecCUDAPlaceArray(ctx->x1,d_px));
29
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 348 PetscCall(VecCUDAPlaceArray(ctx->x2,d_px+m));
30
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 348 PetscCall(VecCUDAPlaceArray(ctx->y1,d_py));
31
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 348 PetscCall(VecCUDAPlaceArray(ctx->y2,d_py+m));
32
2/2
✓ Branch 0 taken 1 times.
✓ Branch 1 taken 1 times.
 348 if (!ctx->misaligned) {
33
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 175 PetscCall(MatMult(ctx->A,ctx->x2,ctx->y1));
34
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 175 PetscCall(MatMult(ctx->AT,ctx->x1,ctx->y2));
35 } else { /* prevent CUDA errors when bottom part is misaligned */
36
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 173 PetscCall(VecCopy(ctx->x2,ctx->wx2));
37
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 173 PetscCall(MatMult(ctx->A,ctx->wx2,ctx->y1));
38
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 173 PetscCall(MatMult(ctx->AT,ctx->x1,ctx->wy2));
39
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 173 PetscCall(VecCopy(ctx->wy2,ctx->y2));
40 }
41
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 348 PetscCall(VecCUDAResetArray(ctx->x1));
42
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 348 PetscCall(VecCUDAResetArray(ctx->x2));
43
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 348 PetscCall(VecCUDAResetArray(ctx->y1));
44
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 348 PetscCall(VecCUDAResetArray(ctx->y2));
45
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 348 PetscCall(VecCUDARestoreArrayRead(x,&d_px));
46
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 348 PetscCall(VecCUDARestoreArrayWrite(y,&d_py));
47 PetscFunctionReturn(PETSC_SUCCESS);
48 }
49
50 4861 PetscErrorCode MatMult_ECross_CUDA(Mat B,Vec x,Vec y)
51 {
52 4861 SVD_CYCLIC_SHELL *ctx;
53 4861 const PetscScalar *d_px;
54 4861 PetscScalar *d_py;
55 4861 PetscInt mn,m,n;
56
57 4861 PetscFunctionBegin;
58
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 4861 PetscCall(MatShellGetContext(B,&ctx));
59
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 4861 PetscCall(MatGetLocalSize(ctx->A,NULL,&n));
60
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 4861 PetscCall(VecGetLocalSize(y,&mn));
61 4861 m = mn-n;
62
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 4861 PetscCall(VecCUDAGetArrayRead(x,&d_px));
63
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 4861 PetscCall(VecCUDAGetArrayWrite(y,&d_py));
64
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 4861 PetscCall(VecCUDAPlaceArray(ctx->x1,d_px));
65
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 4861 PetscCall(VecCUDAPlaceArray(ctx->x2,d_px+m));
66
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 4861 PetscCall(VecCUDAPlaceArray(ctx->y1,d_py));
67
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 4861 PetscCall(VecCUDAPlaceArray(ctx->y2,d_py+m));
68
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 4861 PetscCall(VecCopy(ctx->x1,ctx->y1));
69
2/2
✓ Branch 0 taken 1 times.
✓ Branch 1 taken 1 times.
 4861 if (!ctx->misaligned) {
70
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 2453 PetscCall(MatMult(ctx->A,ctx->x2,ctx->w));
71
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 2453 PetscCall(MatMult(ctx->AT,ctx->w,ctx->y2));
72 } else { /* prevent CUDA errors when bottom part is misaligned */
73
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 2408 PetscCall(VecCopy(ctx->x2,ctx->wx2));
74
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 2408 PetscCall(MatMult(ctx->A,ctx->wx2,ctx->w));
75
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 2408 PetscCall(MatMult(ctx->AT,ctx->w,ctx->wy2));
76
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
 2408 PetscCall(VecCopy(ctx->wy2,ctx->y2));
77 }
78
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 4861 PetscCall(VecCUDAResetArray(ctx->x1));
79
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 4861 PetscCall(VecCUDAResetArray(ctx->x2));
80
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 4861 PetscCall(VecCUDAResetArray(ctx->y1));
81
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 4861 PetscCall(VecCUDAResetArray(ctx->y2));
82
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 4861 PetscCall(VecCUDARestoreArrayRead(x,&d_px));
83
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 4861 PetscCall(VecCUDARestoreArrayWrite(y,&d_py));
84 PetscFunctionReturn(PETSC_SUCCESS);
85 }
86