--- include: rocblas_common.yaml include: known_bugs.yaml Definitions: - &size_t_x_range # - { N: 1073741900, incx: 4 } # high op count: N * incx overflows uint32_t and int32_t - { N: 1048600, incx: 4096 } # low op count: N * incx overflows uint32_t and int32_t - &size_t_xy_range - { N: 1048600, incx: 4096, incy: 1 } - { N: 1048600, incx: 1, incy: 4096 } # TODO add back when workaround for OpenBLAS on windows negative overflow failures #- { N: 1048600, incx: -4096, incy: 1 } #- { N: 1048600, incx: 1, incy: -4096 } - &N_range - [ -1, 0, 5, 10, 500, 1000, 1024, 1025, 7111, 10000, 33792 ] - &incx_range_small - [1, 2] - &incx_range - [-1, 0, 1, 2] - &incx_incy_range - { incx: 1, incy: 1 } - { incx: 1, incy: 2 } - { incx: 2, incy: 1 } - { incx: 1, incy: -1 } - { incx: -1, incy: 1 } - { incx: -3, incy: -3 } - { incx: 0, incy: 1 } - { incx: 0, incy: 0 } # - { incx: 1, incy: 0 } # Indeterminate when y is output - &size_t_pos_x_range - { N: 4, incx: *c_pos_x2_overflow_int32 } # small operation count, should overflow x - &size_t_pos_neg_x_range - { N: 4, incx: *c_pos_x2_overflow_int32 } # small operation count, should overflow x - { N: 4, incx: *c_neg_x2_overflow_int32 } - &size_t_pos_neg_xy_range - { N: 4, incx: *c_pos_x2_overflow_int32, incy: 1 } # small operation count, should overflow x - { N: 4, incx: *c_neg_x2_overflow_int32, incy: 1 } # small operation count, should overflow x - { N: 4, incx: 1, incy: *c_pos_x2_overflow_int32 } # small operation count, should overflow y - { N: 4, incx: 1, incy: *c_neg_x2_overflow_int32 } # small operation count, should overflow y - &incx_incy_range_x_y_output - { incx: 1, incy: 1 } - { incx: 1, incy: 2 } - { incx: 2, incy: 1 } - { incx: 1, incy: -1 } - { incx: -3, incy: 1 } - { incx: -3, incy: -3 } # - { incx: 0, incy: 0 } # Indeterminate for swap # - { incx: 0, incy: 1 } # Indeterminate when x is output (swap) # - { incx: 1, incy: 0 } # Indeterminate when y is output - &incx_incy_range_y_output - { incx: 1, incy: 1 } - { incx: 1, incy: 2 } - { incx: 2, incy: 1 } - { incx: 1, incy: -3 } - { incx: -1, incy: 1 } - { incx: -3, incy: -3 } - { incx: 0, incy: 1 } # - { incx: 0, incy: 0 } # Indeterminate for axpy # - { incx: 1, incy: 0 } # Indeterminate when y is output - &incx_incy_range_small - { incx: 1, incy: 1 } - { incx: -3, incy: -3 } - &alpha_beta_range - { alpha: 1.0, beta: 0.0 } - { alpha: 2.0, beta: -1.0 } - { alpha: 0.0, beta: 0.0 } - &alphai_betai_range - { alphai: 0.0 } - { alphai: 1.0} - { alphai: 2.0 } - &alpha_beta_range_small - { alpha: 0.5, beta: 0.5 } - &complex_alpha_range_small - { alpha: 4.0, alphai: 0.0 } - { alpha: 0.0, alphai: 4.0 } - { alpha: 3.0, alphai: -2.0 } Tests: # for bad_arg no arguments should be used by test code - name: blas1_bad_arg category: quick api: [ C, FORTRAN, FORTRAN_64 ] function: - nrm2_bad_arg: *single_double_precisions_complex_real - nrm2_batched_bad_arg: *single_double_precisions_complex_real - nrm2_strided_batched_bad_arg: *single_double_precisions_complex_real - nrm2_ex_bad_arg: *nrm2_ex_precisions - nrm2_batched_ex_bad_arg: *nrm2_ex_precisions - nrm2_strided_batched_ex_bad_arg: *nrm2_ex_precisions - name: nrm2_ILP64 category: pre_checkin N: [ 64 ] arguments: - { incx: 1, batch_count: 1 } api: [ FORTRAN_64 ] function: - nrm2: *single_double_precisions_complex_real - nrm2_ex: *single_double_precisions_complex_real - nrm2_batched: *single_double_precisions_complex_real - nrm2_batched_ex: *single_double_precisions_complex_real - nrm2_strided_batched: *single_double_precisions_complex_real - nrm2_strided_batched_ex: *single_double_precisions_complex_real - name: nrm2_ILP64 category: stress pointer_mode_host: false initialization: hpl arguments: - { N: 2147483649, incx: 1, batch_count: 1 } - { N: 2, incx: 2147483649, batch_count: 1 } - { N: 2, incx: -2147483649, batch_count: 1 } # quick return - { N: 3, incx: 1, stride_x: 3, batch_count: *c_grid_yz_require_passes } api: [ C_64 ] os_flags: [ LINUX ] function: - nrm2: *single_precision - nrm2_ex: *single_precision - nrm2_batched: *single_precision - nrm2_batched_ex: *single_precision - nrm2_strided_batched: *single_precision - nrm2_strided_batched_ex: *single_precision - name: blas1_size_t category: stress matrix_size: *size_t_pos_x_range pointer_mode_host: false function: - nrm2_ex: *hpa_half_precision gpu_arch: ['90a','942','950'] os_flags: LINUX # All functions with incx, no incy, no alpha # nrm2,asum,iamax,iamin all variants #fortran - name: blas1 category: quick N: [-1, 0, 1025] incx: *incx_range_small api: FORTRAN function: - nrm2: *single_double_precisions_complex_real - nrm2_ex: *nrm2_ex_precisions - name: blas1_batched category: quick N: [-1, 0, 1026] incx: *incx_range_small batch_count: [ -1, 0, 5 ] api: FORTRAN function: - nrm2_batched: *single_double_precisions_complex_real - nrm2_batched_ex: *nrm2_ex_precisions - name: blas1_strided_batched category: quick N: [-1, 0, 1027] incx: *incx_range_small batch_count: [ -1, 0, 5 ] stride_scale: [ 1, 2 ] api: FORTRAN function: - nrm2_strided_batched: *single_double_precisions_complex_real - nrm2_strided_batched_ex: *nrm2_ex_precisions # quick - name: blas1 category: quick N: *N_range incx: *incx_range function: - nrm2: *single_double_precisions_complex_real - nrm2_ex: *nrm2_ex_precisions - name: blas1_batched category: quick N: [ -1, 0, 5, 33792 ] incx: *incx_range batch_count: [-1, 0, 257] function: - nrm2_batched: *single_double_precisions_complex_real - nrm2_batched_ex: *nrm2_ex_precisions - name: blas1_strided_batched category: quick N: [ -1, 0, 5, 33792 ] incx: *incx_range batch_count: [-1, 0, 257] stride_scale: [ 1.5 ] function: - nrm2_strided_batched: *single_double_precisions_complex_real - nrm2_strided_batched_ex: *nrm2_ex_precisions - name: blas1_large_batch category: stress arguments: - { N: 2, incx: 1, stride_x: 2, batch_count: *c_grid_yz_require_passes } function: - nrm2_batched: *single_precision - nrm2_strided_batched: *single_precision - nrm2_batched_ex: *nrm2_ex_precisions_double - nrm2_strided_batched_ex: *nrm2_ex_precisions_double # pre_checkin - name: blas1 category: pre_checkin N: [ 50000, 1049600 ] incx: *incx_range_small function: # for large N nrm2 and asum drop single precision tolerance comparisions. # nrm2_ex uses nrm2 code, no need to test interface again, so only using # the superset code of nrm2_ex. - nrm2_ex: *nrm2_ex_precisions_double - name: blas1_batched category: pre_checkin N: [ 1048576 ] incx: *incx_range_small batch_count: [2] function: - nrm2_batched_ex: *nrm2_ex_precisions_double pointer_mode_host: false - name: blas1_strided_batched category: pre_checkin N: [ 1048576 ] incx: *incx_range_small batch_count: [2] stride_scale: [ 1.5 ] function: - nrm2_strided_batched_ex: *nrm2_ex_precisions_double pointer_mode_device: false # nightly - name: blas1 category: nightly N: [ 300001 ] incx: *incx_range_small function: - nrm2: *double_precision_complex - nrm2_ex: *nrm2_ex_precisions_double - name: blas1_batched category: nightly N: [ 300001 ] incx: *incx_range_small batch_count: [ 65 ] pointer_mode_host: false function: - nrm2_batched_ex: *nrm2_ex_precisions_double - name: blas1_strided_batched category: nightly N: [ 300001 ] incx: *incx_range_small batch_count: [ 65 ] stride_scale: [ 1 ] pointer_mode_device: false function: - nrm2_strided_batched_ex: *nrm2_ex_precisions_double - name: blas1_nan_check category: pre_checkin arguments: *blas1_nan_common_args function: - nrm2: *single_double_precisions_complex_real - nrm2_batched: *single_double_precisions_complex_real - nrm2_strided_batched: *single_double_precisions_complex_real - nrm2_ex: *nrm2_ex_precisions - nrm2_batched_ex: *nrm2_ex_precisions - nrm2_strided_batched_ex: *nrm2_ex_precisions - name: blas1_graph_check category: pre_checkin arguments: *blas1_graph_test_common_args function: - nrm2: *single_double_precisions_complex_real - nrm2_batched: *single_double_precisions_complex_real - nrm2_strided_batched: *single_double_precisions_complex_real - nrm2_ex: *nrm2_ex_precisions - nrm2_batched_ex: *nrm2_ex_precisions - nrm2_strided_batched_ex: *nrm2_ex_precisions - name: blas1_repeatability_check category: stress arguments: *blas1_repeatability_common_args function: - nrm2: *single_double_precisions_complex_real - nrm2_batched: *single_double_precisions_complex_real - nrm2_strided_batched: *single_double_precisions_complex_real - nrm2_ex: *nrm2_ex_precisions - nrm2_batched_ex: *nrm2_ex_precisions - nrm2_strided_batched_ex: *nrm2_ex_precisions ...