diff --git a/cudablas/compute/cuda_zgemerge.c b/cudablas/compute/cuda_zgemerge.c index 5d7aa0cb4b7e0027da4151be5cf0eee98f0279d1..9f3f8d134b1695e69ce1537c429ef013cbd945de 100644 --- a/cudablas/compute/cuda_zgemerge.c +++ b/cudablas/compute/cuda_zgemerge.c @@ -23,12 +23,13 @@ int CUDA_zgemerge( cham_side_t side, cham_diag_t diag, int M, int N, - cuDoubleComplex *A, int LDA, + const cuDoubleComplex *A, int LDA, cuDoubleComplex *B, int LDB, CUBLAS_STREAM_PARAM) { int i; - cuDoubleComplex *cola, *colb; + const cuDoubleComplex *cola; + cuDoubleComplex *colb; if (M < 0) { return -1; diff --git a/cudablas/compute/cuda_zgemm.c b/cudablas/compute/cuda_zgemm.c index ebf104209b20ab1f2a557cd398be0a342a973e9f..7a5900b761ad081181e6b47977e540aa16797b1b 100644 --- a/cudablas/compute/cuda_zgemm.c +++ b/cudablas/compute/cuda_zgemm.c @@ -22,10 +22,10 @@ int CUDA_zgemm(cham_trans_t transa, cham_trans_t transb, int m, int n, int k, - cuDoubleComplex *alpha, + const cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb, - cuDoubleComplex *beta, + const cuDoubleComplex *beta, cuDoubleComplex *C, int ldc, CUBLAS_STREAM_PARAM) { diff --git a/cudablas/compute/cuda_zhemm.c b/cudablas/compute/cuda_zhemm.c index 0a71c08fee4e047293488e3cda871c42b5eaf867..26a470b3414aed3777a23e526ccd8e30febe75b4 100644 --- a/cudablas/compute/cuda_zhemm.c +++ b/cudablas/compute/cuda_zhemm.c @@ -22,10 +22,10 @@ int CUDA_zhemm(cham_side_t side, cham_uplo_t uplo, int m, int n, - cuDoubleComplex *alpha, + const cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb, - cuDoubleComplex *beta, + const cuDoubleComplex *beta, cuDoubleComplex *C, int ldc, CUBLAS_STREAM_PARAM) { diff --git a/cudablas/compute/cuda_zher2k.c b/cudablas/compute/cuda_zher2k.c index b5a8945c75fac27f4df2c54c4eb7fa0a60060c5c..0fd08570d9722bbdcc9a0f419008f343355077a7 100644 --- a/cudablas/compute/cuda_zher2k.c +++ b/cudablas/compute/cuda_zher2k.c @@ -22,10 +22,10 @@ int CUDA_zher2k(cham_uplo_t uplo, cham_trans_t trans, int n, int k, - cuDoubleComplex *alpha, + const cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb, - double *beta, + const double *beta, cuDoubleComplex *C, int ldc, CUBLAS_STREAM_PARAM) { diff --git a/cudablas/compute/cuda_zherk.c b/cudablas/compute/cuda_zherk.c index 34674e9ff31a929f52fc7616e2b6d99dff52d32a..cd0d7ff6441932487e13e5ac4fc6ac9f59d3a12a 100644 --- a/cudablas/compute/cuda_zherk.c +++ b/cudablas/compute/cuda_zherk.c @@ -22,9 +22,9 @@ int CUDA_zherk( cham_uplo_t uplo, cham_trans_t trans, int n, int k, - double *alpha, + const double *alpha, const cuDoubleComplex *A, int lda, - double *beta, + const double *beta, cuDoubleComplex *B, int ldb, CUBLAS_STREAM_PARAM) { diff --git a/cudablas/compute/cuda_zlarfb.c b/cudablas/compute/cuda_zlarfb.c index 51bbbc14601e1414a1d8607cf07a2c8ab892d7cb..3cc573ff59c813e7f9448ac847684e84de4a3f45 100644 --- a/cudablas/compute/cuda_zlarfb.c +++ b/cudablas/compute/cuda_zlarfb.c @@ -42,7 +42,8 @@ CUDA_zlarfb( cham_side_t side, cham_trans_t trans, double mzone = -1.0; #endif /* defined(PRECISION_z) || defined(PRECISION_c) */ - cham_trans_t transT, uplo, notransV, transV; + cham_trans_t transT, notransV, transV; + cham_uplo_t uplo; /* Check input arguments */ if ((side != ChamLeft) && (side != ChamRight)) { diff --git a/cudablas/compute/cuda_zsymm.c b/cudablas/compute/cuda_zsymm.c index ecf201b2585e129f890a0d544564575ce4c1293f..2ed61af73945442e9586c0c2b5a98fb909aecd2a 100644 --- a/cudablas/compute/cuda_zsymm.c +++ b/cudablas/compute/cuda_zsymm.c @@ -20,14 +20,14 @@ */ #include "cudablas.h" -int CUDA_zsymm(cham_side_t side, cham_uplo_t uplo, - int m, int n, - cuDoubleComplex *alpha, - const cuDoubleComplex *A, int lda, - const cuDoubleComplex *B, int ldb, - cuDoubleComplex *beta, - cuDoubleComplex *C, int ldc, - CUBLAS_STREAM_PARAM) +int CUDA_zsymm( cham_side_t side, cham_uplo_t uplo, + int m, int n, + const cuDoubleComplex *alpha, + const cuDoubleComplex *A, int lda, + const cuDoubleComplex *B, int ldb, + const cuDoubleComplex *beta, + cuDoubleComplex *C, int ldc, + CUBLAS_STREAM_PARAM ) { cublasZsymm(CUBLAS_HANDLE chameleon_cublas_const(side), chameleon_cublas_const(uplo), diff --git a/cudablas/compute/cuda_zsyr2k.c b/cudablas/compute/cuda_zsyr2k.c index fb4217d0a09574630dc7f393890547e718fc0840..5c40f1d54fa8034218c9ccd9afd6c409e493186b 100644 --- a/cudablas/compute/cuda_zsyr2k.c +++ b/cudablas/compute/cuda_zsyr2k.c @@ -23,10 +23,10 @@ int CUDA_zsyr2k( cham_uplo_t uplo, cham_trans_t trans, int n, int k, - cuDoubleComplex *alpha, + const cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb, - cuDoubleComplex *beta, + const cuDoubleComplex *beta, cuDoubleComplex *C, int ldc, CUBLAS_STREAM_PARAM) { diff --git a/cudablas/compute/cuda_zsyrk.c b/cudablas/compute/cuda_zsyrk.c index 73a78280606cfea3befcd86e4247fd496bc2867d..b2a61353c747a430b887a661f588f502d713fe79 100644 --- a/cudablas/compute/cuda_zsyrk.c +++ b/cudablas/compute/cuda_zsyrk.c @@ -22,9 +22,9 @@ int CUDA_zsyrk(cham_uplo_t uplo, cham_trans_t trans, int n, int k, - cuDoubleComplex *alpha, + const cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, - cuDoubleComplex *beta, + const cuDoubleComplex *beta, cuDoubleComplex *C, int ldc, CUBLAS_STREAM_PARAM) { diff --git a/cudablas/compute/cuda_ztrmm.c b/cudablas/compute/cuda_ztrmm.c index 71d9fa419a102dee69822f1ad0de5bee2ede21dc..a054413e7619e64d215ad4c68c1ef70294394fd5 100644 --- a/cudablas/compute/cuda_ztrmm.c +++ b/cudablas/compute/cuda_ztrmm.c @@ -24,7 +24,7 @@ int CUDA_ztrmm( cham_side_t side, cham_uplo_t uplo, cham_trans_t transa, cham_diag_t diag, int m, int n, - cuDoubleComplex *alpha, + const cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, cuDoubleComplex *B, int ldb, CUBLAS_STREAM_PARAM) diff --git a/cudablas/compute/cuda_ztrsm.c b/cudablas/compute/cuda_ztrsm.c index e2407be7caaf9d12e8d49360e5b96a1239ba883c..5f63d2464c679c239393040f66bb8f5ef3c1df9b 100644 --- a/cudablas/compute/cuda_ztrsm.c +++ b/cudablas/compute/cuda_ztrsm.c @@ -23,7 +23,7 @@ int CUDA_ztrsm(cham_side_t side, cham_uplo_t uplo, cham_trans_t transa, cham_diag_t diag, int m, int n, - cuDoubleComplex *alpha, + const cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, cuDoubleComplex *B, int ldb, CUBLAS_STREAM_PARAM) diff --git a/cudablas/include/cudablas/cudablas_z.h b/cudablas/include/cudablas/cudablas_z.h index 49fcab19c46bbdf46c00fbce149b20c86096e279..2d22e33ca04a2215f76b477ca2a90c4b6c4cb96f 100644 --- a/cudablas/include/cudablas/cudablas_z.h +++ b/cudablas/include/cudablas/cudablas_z.h @@ -25,21 +25,21 @@ * Declarations of cuda kernels - alphabetical order */ int CUDA_zgeadd( cham_trans_t trans, int m, int n, const cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, const cuDoubleComplex *beta, cuDoubleComplex *B, int ldb, CUBLAS_STREAM_PARAM ); -int CUDA_zgemerge( cham_side_t side, cham_diag_t diag, int M, int N, cuDoubleComplex *A, int LDA, cuDoubleComplex *B, int LDB, CUBLAS_STREAM_PARAM ); -int CUDA_zgemm( cham_trans_t transa, cham_trans_t transb, int m, int n, int k, cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb, cuDoubleComplex *beta, cuDoubleComplex *C, int ldc, CUBLAS_STREAM_PARAM ); -int CUDA_zhemm( cham_side_t side, cham_uplo_t uplo, int m, int n, cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb, cuDoubleComplex *beta, cuDoubleComplex *C, int ldc, CUBLAS_STREAM_PARAM ); -int CUDA_zher2k( cham_uplo_t uplo, cham_trans_t trans, int n, int k, cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb, double *beta, cuDoubleComplex *C, int ldc, CUBLAS_STREAM_PARAM ); +int CUDA_zgemerge( cham_side_t side, cham_diag_t diag, int M, int N, const cuDoubleComplex *A, int LDA, cuDoubleComplex *B, int LDB, CUBLAS_STREAM_PARAM ); +int CUDA_zgemm( cham_trans_t transa, cham_trans_t transb, int m, int n, int k, const cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb, const cuDoubleComplex *beta, cuDoubleComplex *C, int ldc, CUBLAS_STREAM_PARAM ); +int CUDA_zhemm( cham_side_t side, cham_uplo_t uplo, int m, int n, const cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb, const cuDoubleComplex *beta, cuDoubleComplex *C, int ldc, CUBLAS_STREAM_PARAM ); +int CUDA_zher2k( cham_uplo_t uplo, cham_trans_t trans, int n, int k, const cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb, const double *beta, cuDoubleComplex *C, int ldc, CUBLAS_STREAM_PARAM ); int CUDA_zherfb( cham_uplo_t uplo, int n, int k, int ib, int nb, const cuDoubleComplex *A, int lda, const cuDoubleComplex *T, int ldt, cuDoubleComplex *C, int ldc, cuDoubleComplex *WORK, int ldwork, CUBLAS_STREAM_PARAM ); -int CUDA_zherk( cham_uplo_t uplo, cham_trans_t trans, int n, int k, double *alpha, const cuDoubleComplex *A, int lda, double *beta, cuDoubleComplex *B, int ldb, CUBLAS_STREAM_PARAM ); +int CUDA_zherk( cham_uplo_t uplo, cham_trans_t trans, int n, int k, const double *alpha, const cuDoubleComplex *A, int lda, const double *beta, cuDoubleComplex *B, int ldb, CUBLAS_STREAM_PARAM ); int CUDA_zlarfb(cham_side_t side, cham_trans_t trans, cham_dir_t direct, cham_store_t storev, int M, int N, int K, const cuDoubleComplex *V, int LDV, const cuDoubleComplex *T, int LDT, cuDoubleComplex *C, int LDC, cuDoubleComplex *WORK, int LDWORK, CUBLAS_STREAM_PARAM ); int CUDA_zparfb(cham_side_t side, cham_trans_t trans, cham_dir_t direct, cham_store_t storev, int M1, int N1, int M2, int N2, int K, int L, cuDoubleComplex *A1, int LDA1, cuDoubleComplex *A2, int LDA2, const cuDoubleComplex *V, int LDV, const cuDoubleComplex *T, int LDT, cuDoubleComplex *WORK, int LWORK, CUBLAS_STREAM_PARAM ); -int CUDA_zsymm( cham_side_t side, cham_uplo_t uplo, int m, int n, cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb, cuDoubleComplex *beta, cuDoubleComplex *C, int ldc, CUBLAS_STREAM_PARAM ); -int CUDA_zsyr2k( cham_uplo_t uplo, cham_trans_t trans, int n, int k, cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb, cuDoubleComplex *beta, cuDoubleComplex *C, int ldc, CUBLAS_STREAM_PARAM ); -int CUDA_zsyrk( cham_uplo_t uplo, cham_trans_t trans, int n, int k, cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, cuDoubleComplex *beta, cuDoubleComplex *C, int ldc, CUBLAS_STREAM_PARAM ); +int CUDA_zsymm( cham_side_t side, cham_uplo_t uplo, int m, int n, const cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb, const cuDoubleComplex *beta, cuDoubleComplex *C, int ldc, CUBLAS_STREAM_PARAM ); +int CUDA_zsyr2k( cham_uplo_t uplo, cham_trans_t trans, int n, int k, const cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, const cuDoubleComplex *B, int ldb, const cuDoubleComplex *beta, cuDoubleComplex *C, int ldc, CUBLAS_STREAM_PARAM ); +int CUDA_zsyrk( cham_uplo_t uplo, cham_trans_t trans, int n, int k, const cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, const cuDoubleComplex *beta, cuDoubleComplex *C, int ldc, CUBLAS_STREAM_PARAM ); int CUDA_ztpmqrt( cham_side_t side, cham_trans_t trans, int M, int N, int K, int L, int IB, const cuDoubleComplex *V, int LDV, const cuDoubleComplex *T, int LDT, cuDoubleComplex *A, int LDA, cuDoubleComplex *B, int LDB, cuDoubleComplex *WORK, int lwork, CUBLAS_STREAM_PARAM ); int CUDA_ztpmlqt( cham_side_t side, cham_trans_t trans, int M, int N, int K, int L, int IB, const cuDoubleComplex *V, int LDV, const cuDoubleComplex *T, int LDT, cuDoubleComplex *A, int LDA, cuDoubleComplex *B, int LDB, cuDoubleComplex *WORK, int lwork, CUBLAS_STREAM_PARAM ); -int CUDA_ztrmm( cham_side_t side, cham_uplo_t uplo, cham_trans_t transa, cham_diag_t diag, int m, int n, cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, cuDoubleComplex *B, int ldb, CUBLAS_STREAM_PARAM ); -int CUDA_ztrsm( cham_side_t side, cham_uplo_t uplo, cham_trans_t transa, cham_diag_t diag, int m, int n, cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, cuDoubleComplex *B, int ldb, CUBLAS_STREAM_PARAM ); +int CUDA_ztrmm( cham_side_t side, cham_uplo_t uplo, cham_trans_t transa, cham_diag_t diag, int m, int n, const cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, cuDoubleComplex *B, int ldb, CUBLAS_STREAM_PARAM ); +int CUDA_ztrsm( cham_side_t side, cham_uplo_t uplo, cham_trans_t transa, cham_diag_t diag, int m, int n, const cuDoubleComplex *alpha, const cuDoubleComplex *A, int lda, cuDoubleComplex *B, int ldb, CUBLAS_STREAM_PARAM ); int CUDA_ztsmlq( cham_side_t side, cham_trans_t trans, int M1, int N1, int M2, int N2, int K, int IB, cuDoubleComplex *A1, int LDA1, cuDoubleComplex *A2, int LDA2, const cuDoubleComplex *V, int LDV, const cuDoubleComplex *T, int LDT, cuDoubleComplex *WORK, int LWORK, CUBLAS_STREAM_PARAM ); int CUDA_zttmlq( cham_side_t side, cham_trans_t trans, int M1, int N1, int M2, int N2, int K, int IB, cuDoubleComplex *A1, int LDA1, cuDoubleComplex *A2, int LDA2, const cuDoubleComplex *V, int LDV, const cuDoubleComplex *T, int LDT, cuDoubleComplex *WORK, int LWORK, CUBLAS_STREAM_PARAM ); int CUDA_ztsmqr( cham_side_t side, cham_trans_t trans, int M1, int N1, int M2, int N2, int K, int IB, cuDoubleComplex *A1, int LDA1, cuDoubleComplex *A2, int LDA2, const cuDoubleComplex *V, int LDV, const cuDoubleComplex *T, int LDT, cuDoubleComplex *WORK, int LWORK, CUBLAS_STREAM_PARAM );