/***************************************************************************//** Purpose ------- CUNMTR overwrites the general complex M-by-N matrix C with SIDE = MagmaLeft SIDE = MagmaRight TRANS = MagmaNoTrans: Q * C C * Q TRANS = Magma_ConjTrans: Q**H * C C * Q**H where Q is a complex unitary matrix of order nq, with nq = m if SIDE = MagmaLeft and nq = n if SIDE = MagmaRight. Q is defined as the product of nq-1 elementary reflectors, as returned by CHETRD: if UPLO = MagmaUpper, Q = H(nq-1) . . . H(2) H(1); if UPLO = MagmaLower, Q = H(1) H(2) . . . H(nq-1). Arguments --------- @param[in] side magma_side_t - = MagmaLeft: apply Q or Q**H from the Left; - = MagmaRight: apply Q or Q**H from the Right. @param[in] uplo magma_uplo_t - = MagmaUpper: Upper triangle of A contains elementary reflectors from CHETRD; - = MagmaLower: Lower triangle of A contains elementary reflectors from CHETRD. @param[in] trans magma_trans_t - = MagmaNoTrans: No transpose, apply Q; - = Magma_ConjTrans: Conjugate transpose, apply Q**H. @param[in] m INTEGER The number of rows of the matrix C. M >= 0. @param[in] n INTEGER The number of columns of the matrix C. N >= 0. @param[in] A COMPLEX array, dimension (LDA,M) if SIDE = MagmaLeft (LDA,N) if SIDE = MagmaRight The vectors which define the elementary reflectors, as returned by CHETRD. @param[in] lda INTEGER The leading dimension of the array A. LDA >= max(1,M) if SIDE = MagmaLeft; LDA >= max(1,N) if SIDE = MagmaRight. @param[in] tau COMPLEX array, dimension (M-1) if SIDE = MagmaLeft (N-1) if SIDE = MagmaRight TAU(i) must contain the scalar factor of the elementary reflector H(i), as returned by CHETRD. @param[in,out] C COMPLEX array, dimension (LDC,N) On entry, the M-by-N matrix C. On exit, C is overwritten by Q*C or Q**H * C or C * Q**H or C*Q. @param[in] ldc INTEGER The leading dimension of the array C. LDC >= max(1,M). @param[out] work (workspace) COMPLEX array, dimension (MAX(1,LWORK)) On exit, if INFO = 0, WORK[0] returns the optimal LWORK. @param[in] lwork INTEGER The dimension of the array WORK. If SIDE = MagmaLeft, LWORK >= max(1,N); if SIDE = MagmaRight, LWORK >= max(1,M). For optimum performance LWORK >= N*NB if SIDE = MagmaLeft, and LWORK >= M*NB if SIDE = MagmaRight, where NB is the optimal blocksize. \n If LWORK = -1, then a workspace query is assumed; the routine only calculates the optimal size of the WORK array, returns this value as the first entry of the WORK array, and no error message related to LWORK is issued. @param[out] info INTEGER - = 0: successful exit - < 0: if INFO = -i, the i-th argument had an illegal value @ingroup magma_unmtr *******************************************************************************/ extern "C" magma_int_t magma_cunmtr( magma_side_t side, magma_uplo_t uplo, magma_trans_t trans, magma_int_t m, magma_int_t n, magmaFloatComplex *A, magma_int_t lda, magmaFloatComplex *tau, magmaFloatComplex *C, magma_int_t ldc, magmaFloatComplex *work, magma_int_t lwork, magma_int_t *info) { #define A(i_,j_) (A + (i_) + (j_)*lda) #define C(i_,j_) (C + (i_) + (j_)*ldc) magmaFloatComplex c_one = MAGMA_C_ONE; magma_int_t i__2; magma_int_t i1, i2, nb, mi, ni, nq, nw; magma_int_t iinfo; magma_int_t lwkopt; *info = 0; bool left = (side == MagmaLeft); bool upper = (uplo == MagmaUpper); bool lquery = (lwork == -1); /* NQ is the order of Q and NW is the minimum dimension of WORK */ if (left) { nq = m; nw = n; } else { nq = n; nw = m; } if (! left && side != MagmaRight) { *info = -1; } else if (! upper && uplo != MagmaLower) { *info = -2; } else if (trans != MagmaNoTrans && trans != Magma_ConjTrans) { *info = -3; } else if (m < 0) { *info = -4; } else if (n < 0) { *info = -5; } else if (lda < max(1,nq)) { *info = -7; } else if (ldc < max(1,m)) { *info = -10; } else if (lwork < max(1,nw) && ! lquery) { *info = -12; } nb = 32; lwkopt = max(1,nw) * nb; if (*info == 0) { work[0] = magma_cmake_lwork( lwkopt ); } if (*info != 0) { magma_xerbla( __func__, -(*info) ); return *info; } else if (lquery) { return *info; } /* Quick return if possible */ if (m == 0 || n == 0 || nq == 1) { work[0] = c_one; return *info; } if (left) { mi = m - 1; ni = n; } else { mi = m; ni = n - 1; } if (upper) { /* Q was determined by a call to CHETRD with UPLO = MagmaUpper */ i__2 = nq - 1; //lapackf77_cunmql(side_, trans_, &mi, &ni, &i__2, A(0,1), &lda, // tau, C, &ldc, work, &lwork, &iinfo); magma_cunmql(side, trans, mi, ni, i__2, A(0,1), lda, tau, C, ldc, work, lwork, &iinfo); } else { /* Q was determined by a call to CHETRD with UPLO = MagmaLower */ if (left) { i1 = 1; i2 = 0; } else { i1 = 0; i2 = 1; } i__2 = nq - 1; magma_cunmqr(side, trans, mi, ni, i__2, A(1,0), lda, tau, C(i1,i2), ldc, work, lwork, &iinfo); } work[0] = magma_cmake_lwork( lwkopt ); return *info; } /* magma_cunmtr */
/* //////////////////////////////////////////////////////////////////////////// -- Testing cunmql */ int main( int argc, char** argv ) { TESTING_INIT(); real_Double_t gflops, gpu_perf, gpu_time, cpu_perf, cpu_time; float error, work[1]; magmaFloatComplex c_neg_one = MAGMA_C_NEG_ONE; magma_int_t ione = 1; magma_int_t mm, m, n, k, size, info; magma_int_t ISEED[4] = {0,0,0,1}; magma_int_t nb, ldc, lda, lwork, lwork_max; magmaFloatComplex *C, *R, *A, *W, *tau; magma_int_t status = 0; magma_opts opts; parse_opts( argc, argv, &opts ); // need slightly looser bound (60*eps instead of 30*eps) for some tests opts.tolerance = max( 60., opts.tolerance ); float tol = opts.tolerance * lapackf77_slamch("E"); // test all combinations of input parameters magma_side_t side [] = { MagmaLeft, MagmaRight }; magma_trans_t trans[] = { Magma_ConjTrans, MagmaNoTrans }; printf(" M N K side trans CPU GFlop/s (sec) GPU GFlop/s (sec) ||R||_F / ||QC||_F\n"); printf("===============================================================================================\n"); for( int itest = 0; itest < opts.ntest; ++itest ) { for( int iside = 0; iside < 2; ++iside ) { for( int itran = 0; itran < 2; ++itran ) { for( int iter = 0; iter < opts.niter; ++iter ) { m = opts.msize[itest]; n = opts.nsize[itest]; k = opts.ksize[itest]; nb = magma_get_cgeqlf_nb( m ); ldc = m; // A is m x k (left) or n x k (right) mm = (side[iside] == MagmaLeft ? m : n); lda = mm; gflops = FLOPS_CUNMQL( m, n, k, side[iside] ) / 1e9; if ( side[iside] == MagmaLeft && m < k ) { printf( "%5d %5d %5d %4c %5c skipping because side=left and m < k\n", (int) m, (int) n, (int) k, lapacke_side_const( side[iside] ), lapacke_trans_const( trans[itran] ) ); continue; } if ( side[iside] == MagmaRight && n < k ) { printf( "%5d %5d %5d %4c %5c skipping because side=right and n < k\n", (int) m, (int) n, (int) k, lapacke_side_const( side[iside] ), lapacke_trans_const( trans[itran] ) ); continue; } // need at least 2*nb*nb for geqlf lwork_max = max( max( m*nb, n*nb ), 2*nb*nb ); TESTING_MALLOC_CPU( C, magmaFloatComplex, ldc*n ); TESTING_MALLOC_CPU( R, magmaFloatComplex, ldc*n ); TESTING_MALLOC_CPU( A, magmaFloatComplex, lda*k ); TESTING_MALLOC_CPU( W, magmaFloatComplex, lwork_max ); TESTING_MALLOC_CPU( tau, magmaFloatComplex, k ); // C is full, m x n size = ldc*n; lapackf77_clarnv( &ione, ISEED, &size, C ); lapackf77_clacpy( "Full", &m, &n, C, &ldc, R, &ldc ); size = lda*k; lapackf77_clarnv( &ione, ISEED, &size, A ); // compute QL factorization to get Householder vectors in A, tau magma_cgeqlf( mm, k, A, lda, tau, W, lwork_max, &info ); if (info != 0) printf("magma_cgeqlf returned error %d: %s.\n", (int) info, magma_strerror( info )); /* ===================================================================== Performs operation using LAPACK =================================================================== */ cpu_time = magma_wtime(); lapackf77_cunmql( lapack_side_const( side[iside] ), lapack_trans_const( trans[itran] ), &m, &n, &k, A, &lda, tau, C, &ldc, W, &lwork_max, &info ); cpu_time = magma_wtime() - cpu_time; cpu_perf = gflops / cpu_time; if (info != 0) printf("lapackf77_cunmql returned error %d: %s.\n", (int) info, magma_strerror( info )); /* ==================================================================== Performs operation using MAGMA =================================================================== */ // query for workspace size lwork = -1; magma_cunmql( side[iside], trans[itran], m, n, k, A, lda, tau, R, ldc, W, lwork, &info ); if (info != 0) printf("magma_cunmql (lwork query) returned error %d: %s.\n", (int) info, magma_strerror( info )); lwork = (magma_int_t) MAGMA_C_REAL( W[0] ); if ( lwork < 0 || lwork > lwork_max ) { printf("optimal lwork %d > lwork_max %d\n", (int) lwork, (int) lwork_max ); lwork = lwork_max; } gpu_time = magma_wtime(); magma_cunmql( side[iside], trans[itran], m, n, k, A, lda, tau, R, ldc, W, lwork, &info ); gpu_time = magma_wtime() - gpu_time; gpu_perf = gflops / gpu_time; if (info != 0) printf("magma_cunmql returned error %d: %s.\n", (int) info, magma_strerror( info )); /* ===================================================================== compute relative error |QC_magma - QC_lapack| / |QC_lapack| =================================================================== */ error = lapackf77_clange( "Fro", &m, &n, C, &ldc, work ); size = ldc*n; blasf77_caxpy( &size, &c_neg_one, C, &ione, R, &ione ); error = lapackf77_clange( "Fro", &m, &n, R, &ldc, work ) / error; printf( "%5d %5d %5d %4c %5c %7.2f (%7.2f) %7.2f (%7.2f) %8.2e %s\n", (int) m, (int) n, (int) k, lapacke_side_const( side[iside] ), lapacke_trans_const( trans[itran] ), cpu_perf, cpu_time, gpu_perf, gpu_time, error, (error < tol ? "ok" : "failed") ); status += ! (error < tol); TESTING_FREE_CPU( C ); TESTING_FREE_CPU( R ); TESTING_FREE_CPU( A ); TESTING_FREE_CPU( W ); TESTING_FREE_CPU( tau ); fflush( stdout ); } if ( opts.niter > 1 ) { printf( "\n" ); } }} // end iside, itran printf( "\n" ); } TESTING_FINALIZE(); return status; }