--- 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: # All functions with no parameters to scan over # for bad_arg no arguments should be used by test code - name: blas1_bad_arg category: quick api: [ C, FORTRAN, FORTRAN_64 ] function: - rotg_bad_arg: *rotg_precisions - rotm_bad_arg: *single_double_precisions - rotmg_bad_arg: *single_double_precisions - rotg_batched_bad_arg: *rotg_precisions - rotm_batched_bad_arg: *single_double_precisions - rotmg_batched_bad_arg: *single_double_precisions - rotg_strided_batched_bad_arg: *rotg_precisions - rotm_strided_batched_bad_arg: *single_double_precisions - rotmg_strided_batched_bad_arg: *single_double_precisions - rot_bad_arg: *rot_precisions - rot_ex_bad_arg: *rot_ex_precisions - rot_batched_bad_arg: *rot_precisions - rot_batched_ex_bad_arg: *rot_ex_precisions - rot_strided_batched_bad_arg: *rot_precisions - rot_strided_batched_ex_bad_arg: *rot_ex_precisions - name: blas1 category: quick api: [ C, FORTRAN ] function: - rotg: *rotg_precisions - rotg_batched: *rotg_precisions - rotg_strided_batched: *rotg_precisions alpha_beta: # using alpha and beta in place of a and b - { alpha: 0.0, beta: 0.0 , alphai: 0.0, betai: 0.0} # using alpha and beta in place of a and b - { alpha: 1.0, beta: 2.0 , alphai: 0.0, betai: 0.0} - { alpha: 1.1, beta: 0.0 , alphai: 0.1, betai: 0.0} - { alpha: 0.0, beta: 1.2 , alphai: 0.0, betai: 0.2} - { alpha: -1.3, beta: 1.4 , alphai: 0.3, betai: -0.4} - { alpha: 1.6, beta: -1.5 , alphai: -0.6, betai: 0.5} - name: blas1 category: quick api: [ C, FORTRAN ] function: - rotg: *rotg_precisions - rotmg: *single_double_precisions - name: blas1_batched category: quick batch_count: [-1, 0, 5] api: [ C, FORTRAN ] function: - rotg_batched: *rotg_precisions - rotmg_batched: *single_double_precisions - name: blas1_strided_batched category: quick batch_count: [-1, 0, 5] stride_scale: [ 1.5 ] api: [ C, FORTRAN ] function: - rotg_strided_batched: *rotg_precisions - rotmg_strided_batched: *single_double_precisions - name: rot_ILP64 category: pre_checkin N: [ 64 ] arguments: - { incx: 1, incy: 1, batch_count: 1 } api: [ FORTRAN_64 ] function: - rot: *rot_precisions - rot_ex: *rot_ex_precisions - rot_batched: *rot_precisions - rot_batched_ex: *rot_ex_precisions - rot_strided_batched: *rot_precisions - rot_strided_batched_ex: *rot_ex_precisions - name: rot_ILP64 category: stress arguments: - { N: 2147483649, incx: 1, incy: 1, batch_count: 1 } - { N: 2, incx: 2147483649, incy: 1, batch_count: 1 } - { N: 2, incx: 1, incy: 2147483649, batch_count: 1 } - { N: 3, incx: 1, incy: 1, stride_x: 3, stride_y: 3, batch_count: *c_grid_yz_require_passes } - { N: 2, incx: -2147483649, incy: 1, batch_count: 1 } - { N: 2, incx: 1, incy: -2147483649, batch_count: 1 } api: [ C_64 ] os_flags: [ LINUX ] function: - rot: *single_precision - rot_ex: *single_precision - rot_batched: *single_precision - rot_batched_ex: *single_precision - rot_strided_batched: *single_precision - rot_strided_batched_ex: *single_precision - name: blas1_size_t category: stress matrix_size: *size_t_xy_range pointer_mode_device: false function: - rot_ex: *hpa_half_precision # real rot shared kernels with rot_ex so only testing size_t with rot_ex and half. gpu_arch: ['90a','942','950'] os_flags: LINUX - name: blas1_size_t category: stress matrix_size: *size_t_xy_range pointer_mode_device: false function: - rot_ex: *single_precision_complex # complex rot shared kernels with rot_ex so only testing size_t with rot_ex and single_precision_complex. gpu_arch: ['90a','942','950'] os_flags: LINUX - name: blas1_size_t category: stress matrix_size: *size_t_xy_range function: - rotm: *single_precision #all rotm functions use same templated source so it is only necessary to test single_precision. gpu_arch: ['90a','942','950'] os_flags: LINUX # All functions with incx, incy, no alpha # fortran - name: blas1 category: quick N: [ -1, 0, 1025] incx_incy: *incx_incy_range_small api: [FORTRAN, C_64] function: - rot: *rot_precisions - rot_ex: *rot_ex_precisions - rotm: *single_double_precisions - name: blas1_large_batch category: stress N: [ 5 ] incx: 1 incy: 1 batch_count: [ *c_grid_yz_require_passes ] stride_scale: [ 1 ] function: - rot_batched: *rot_precisions - rot_batched_ex: *rot_ex_precisions - rot_strided_batched: *rot_precisions - rot_strided_batched_ex: *rot_ex_precisions - rotm_batched: *single_double_precisions - rotm_strided_batched: *single_double_precisions - name: blas1_batched category: quick N: [ -1, 0, 1025] incx_incy: *incx_incy_range_small batch_count: [-1, 0, 5] api: FORTRAN function: - rot_batched: *rot_precisions - rot_batched_ex: *rot_ex_precisions - rotm_batched: *single_double_precisions - name: blas1_strided_batched category: quick N: [ -1, 0, 1025] incx_incy: *incx_incy_range_small batch_count: [-1, 0, 5] stride_scale: [1, 2] api: FORTRAN function: - rot_strided_batched: *rot_precisions - rot_strided_batched_ex: *rot_ex_precisions - rotm_strided_batched: *single_double_precisions - name: blas1 category: pre_checkin N: [ 5000, 800000 ] pointer_mode_host: false incx_incy: *incx_incy_range_small function: # dot_ex uses dot code, so only testing superset code of dot_ex for pre-checkin and nightly. # we should be sure to keep testing the dot API in quick tests as done above. # similarly with rot - rot_ex: *rot_ex_precisions - rotm: *single_double_precisions - name: blas1_batched category: pre_checkin N: [ 5000, 800000 ] pointer_mode_device: false incx_incy: *incx_incy_range_small batch_count: [1, 7] function: - rot_batched_ex: *rot_ex_precisions - rotm_batched: *single_double_precisions - name: blas1_strided_batched category: pre_checkin N: [ 5000, 800000 ] pointer_mode_host: false incx_incy: *incx_incy_range_small batch_count: [1, 7] stride_scale: [ 1 ] function: - rot_strided_batched_ex: *rot_ex_precisions - rotm_strided_batched: *single_double_precisions # nightly - name: blas1 category: nightly N: [ 1000000 ] pointer_mode_device: false incx_incy: *incx_incy_range_small function: - rot_ex: *rot_ex_precisions - rotm: *single_double_precisions - name: blas1_batched category: nightly N: [ 1000000 ] pointer_mode_host: false incx_incy: *incx_incy_range_small batch_count: [1, 7] function: - rot_batched_ex: *rot_ex_precisions - rotm_batched: *single_double_precisions - name: blas1_strided_batched category: nightly N: [ 1000000 ] pointer_mode_device: false incx_incy: *incx_incy_range_small batch_count: [1, 7] stride_scale: [ 1 ] function: - rot_strided_batched_ex: *rot_ex_precisions - rotm_strided_batched: *single_double_precisions # Swap, rot, and rotm get their own for quick tests as they write to x and y # quick - name: blas1 category: quick N: [ -1, 0, 32, 1025] incx_incy: *incx_incy_range_x_y_output batch_count: [1, 7] stride_scale: [ 1 ] function: - rot: *rot_precisions - rot_batched: *rot_precisions - rot_strided_batched: *rot_precisions - rot_ex: *rot_ex_precisions - rot_batched_ex: *rot_ex_precisions - rot_strided_batched_ex: *rot_ex_precisions - rotm: *single_double_precisions - rotm_batched: *single_double_precisions - rotm_strided_batched: *single_double_precisions # quick tests will fallback to original kernels at launcher level - name: rotm_ILP64 category: nightly N: [ 64 ] incx_incy: *incx_incy_range_small api: [ FORTRAN_64 ] function: - rotm: *single_double_precisions_complex_real - rotm_batched: *single_double_precisions_complex_real - rotm_strided_batched: *single_double_precisions_complex_real - name: rotmg_ILP64 category: nightly api: [ FORTRAN_64 ] function: - rotmg: *single_double_precisions_complex_real - rotmg_batched: *single_double_precisions_complex_real - rotmg_strided_batched: *single_double_precisions_complex_real - name: rotg_ILP64 category: nightly api: [ FORTRAN_64 ] function: - rotg: *single_double_precisions_complex_real - rotg_batched: *single_double_precisions_complex_real - rotg_strided_batched: *single_double_precisions_complex_real - name: rotm_ILP64 category: stress arguments: # stride_c is param stride so must be >= 5 - { N: 2147483649, incx: 1, incy: -1, stride_x: 2, stride_y: 2, stride_c: 5, batch_count: 1 } - { N: 2, incx: -2147483649, incy: 1, stride_x: 2, stride_y: 2, stride_c: 5, batch_count: 1 } - { N: 2, incx: -1, incy: 1, stride_x: 2, stride_y: 2, stride_c: 5, batch_count: *c_grid_yz_require_passes } api: [ C_64 ] os_flags: [ LINUX ] pointer_mode_host: false function: - rotm: *single_precision - rotm_batched: *single_precision - rotm_strided_batched: *single_precision - name: rotmg_ILP64 category: stress arguments: # stride_c is param stride so must be >= 5 - { stride_a: 1, stride_b: 1, stride_x: 1, stride_y: 1, stride_c: 5, batch_count: *c_grid_yz_require_passes } api: [ C_64 ] os_flags: [ LINUX ] pointer_mode_host: false function: - rotmg: *double_precision - rotmg_batched: *single_precision - rotmg_strided_batched: *single_precision - name: rotg_ILP64 category: stress arguments: - { stride_a: 1, stride_b: 1, stride_c: 1, stride_d: 1, batch_count: *c_grid_yz_require_passes } api: [ C_64 ] os_flags: [ LINUX ] pointer_mode_host: false function: - rotg: *double_precision - rotg_batched: *single_precision - rotg_strided_batched: *single_precision - name: blas1_nan_check category: pre_checkin arguments: *blas1_nan_common_args function: #TODO: - Tests bypass the unit_check_general in ROTG,ROTMG,ROTM when NaN is input for alpha as this is not expected - rot: *rot_precisions - rot_ex: *rot_ex_precisions - rotm: *single_double_precisions - rotmg: *single_double_precisions - rot_batched: *rot_precisions - rot_batched_ex: *rot_ex_precisions - rotm_batched: *single_double_precisions - rotmg_batched: *single_double_precisions - rot_strided_batched: *rot_precisions - rot_strided_batched_ex: *rot_ex_precisions - rotm_strided_batched: *single_double_precisions - rotmg_strided_batched: *single_double_precisions - name: blas1_graph_check category: pre_checkin arguments: *blas1_graph_test_common_args function: - rot: *rot_precisions - rot_ex: *rot_ex_precisions - rotg: *rotg_precisions - rotm: *single_double_precisions - rotmg: *single_double_precisions - rot_batched: *rot_precisions - rot_batched_ex: *rot_ex_precisions - rotg_batched: *rotg_precisions - rotm_batched: *single_double_precisions - rotmg_batched: *single_double_precisions - rot_strided_batched: *rot_precisions - rotg_strided_batched: *rotg_precisions - rotm_strided_batched: *single_double_precisions - rotmg_strided_batched: *single_double_precisions - name: blas1_repeatability_check category: stress arguments: *blas1_repeatability_common_args function: - rot: *rot_precisions - rot_ex: *rot_ex_precisions - rotg: *rotg_precisions - rotm: *single_double_precisions - rotmg: *single_double_precisions - rot_batched: *rot_precisions - rot_batched_ex: *rot_ex_precisions - rotg_batched: *rotg_precisions - rotm_batched: *single_double_precisions - rotmg_batched: *single_double_precisions - rot_strided_batched: *rot_precisions - rotg_strided_batched: *rotg_precisions - rotm_strided_batched: *single_double_precisions - rotmg_strided_batched: *single_double_precisions ...