---
include: rocblas_common.yaml
include: known_bugs.yaml

Definitions:
  - &invalid_size_range
    # - { M:  3, N:  3, K:  3, lda: 3, ldb: 3, ldc: 3, ldd: 3, stride_scale: 1, transA: N, transB: N, batch_count: -1 } # batch_count < 0
    - { M: -1, N:  1, K:  1, lda: 1, ldb: 1, ldc: 1, ldd: 1, stride_scale: 1, transA: N, transB: N } # M < 0
    - { M:  1, N: -1, K:  1, lda: 1, ldb: 1, ldc: 1, ldd: 1, stride_scale: 1, transA: N, transB: N } # N < 0
    - { M:  1, N:  1, K: -1, lda: 1, ldb: 1, ldc: 1, ldd: 1, stride_scale: 1, transA: N, transB: N } # K < 0
    - { M:  2, N:  1, K:  1, lda: 1, ldb: 2, ldc: 2, ldd: 2, stride_scale: 1, transA: N, transB: N } # lda < M, transA = N
    - { M:  1, N:  1, K:  2, lda: 1, ldb: 2, ldc: 2, ldd: 2, stride_scale: 1, transA: T, transB: N } # lda < K, transA = T
    - { M:  1, N:  1, K:  2, lda: 2, ldb: 1, ldc: 2, ldd: 2, stride_scale: 1, transA: N, transB: N } # ldb < K, transB = N
    - { M:  2, N:  2, K:  1, lda: 2, ldb: 1, ldc: 2, ldd: 2, stride_scale: 1, transA: N, transB: T } # ldb < N, transB = T
    - { M:  2, N:  1, K:  1, lda: 2, ldb: 2, ldc: 1, ldd: 2, stride_scale: 1, transA: N, transB: N } # ldc < M
    - { M:  2, N:  1, K:  1, lda: 2, ldb: 2, ldc: 2, ldd: 1, stride_scale: 1, transA: N, transB: N } # ldd < M

  - &small_matrix_size_range
    - { M:     0, N:     1, K: 1,  lda:     1, ldb:     1, ldc:     1, ldd:   1 } # M == 0
    - { M:     1, N:     0, K: 1,  lda:     1, ldb:     1, ldc:     1, ldd:   1 } # N == 0
    - { M:   256, N:   128, K: 0,  lda:   256, ldb:   256, ldc:   256, ldd: 256 } # K == 0
    - { M:     3, N:    33, K: 15, lda:    35, ldb:    35, ldc:    35, ldd:  85 }
    - { M:    10, N:    11, K: 17, lda:   100, ldb:    12, ldc:    13, ldd:  10 }
    - { M:     3, N:    33, K: 15, lda:    35, ldb:    35, ldc:    35, ldd:  35 }

  - &large_matrix_size_range
    - { M:   192, N:   193, lda:   194, ldb:   195, ldc:   196 }
    - { M:   640, N:   641, lda:   960, ldb:   961, ldc:   961 }
    - { M:  1001, N:  1000, lda:  1003, ldb:  1002, ldc:  1001 }

  - &huge_matrix_size_range
    - { M:  4011, N:  4012, lda:  4012, ldb:  4013, ldc:  4014 }

  - &small_alpha_beta_range
    - { alpha:  0.0, beta:  0.0 }
    - { alpha:  2.0, beta:  0.0 }
    - { alpha:  0.0, beta:  2.0 }
    - { alpha: -3.0, beta: -2.0 }

  - &large_alpha_beta_range
    - { alpha:  2.0, beta:  0.0 }
    - { alpha:  0.0, beta:  2.0 }
    - { alpha: -2.0, beta: -3.0 }

  - &huge_alpha_beta_range
    - { alpha:  1.0, beta:  3.0 }

  - &transA_transB_range
    - { transA: [N,T], transB: [N,T] }

  - &tiny_transA_transB_range
    - { transA: N, transB: N }


Tests:
- name: bad_arg
  category: quick
  function: geam_ex_bad_arg
  precision: *half_single_double_precisions
  geam_op: [0, 1]
  api: [ C, FORTRAN ]

- name: geam_ex_invalid_size
  category: quick
  function: geam_ex
  precision: *single_double_precisions
  matrix_size: *invalid_size_range
  api: [ C, FORTRAN ]
  geam_op: [0, 1]

- name: geam_ex_size_t
  category: stress
  function: geam_ex
  precision: *single_precision
  transA: [ N ]
  transB: [ N ]
  matrix_size:
    - { M: 4, N:     4, K: 51200, lda: 51200, ldb:     4, ldc:     4, ldd:     4 }
    - { M: 4, N: 51200, K:     4, lda:     4, ldb: 51200, ldc:     4, ldd:     4 }
    - { M: 4, N: 51200, K:     4, lda:     4, ldb:     4, ldc: 51200, ldd:     4 }
    - { M: 4, N: 51200, K:     4, lda:     4, ldb:     4, ldc:     4, ldd: 51200}
  api: C
  geam_op: [0, 1]

- name: geam_ex_small
  category: quick
  function: geam_ex
  precision: *half_single_double_precisions
  transA_transB: *transA_transB_range
  matrix_size: *small_matrix_size_range
  alpha_beta: *small_alpha_beta_range
  api: [ C, FORTRAN ]
  geam_op: [0, 1]

- name: geam_ex_large
  category: pre_checkin
  function: geam_ex
  precision: *half_single_double_precisions
  transA_transB: *transA_transB_range
  matrix_size: *large_matrix_size_range
  alpha_beta: *large_alpha_beta_range
  geam_op: [0, 1]

- name: geam_ex_huge
  category: nightly
  function: geam_ex
  precision: *half_single_double_precisions
  transA_transB: *transA_transB_range
  matrix_size: *huge_matrix_size_range
  alpha_beta: *huge_alpha_beta_range
  geam_op: [0, 1]

- name: geam_ex_graph_test
  category: pre_checkin
  function: geam_ex
  precision: *half_single_double_precisions
  transA_transB: *tiny_transA_transB_range
  matrix_size:
    -  { M:     3, N:    33, K: 15, lda:    35, ldb:    35, ldc:    35, ldd:  85 }
  alpha_beta: *small_alpha_beta_range
  geam_op: [0, 1]
  graph_test: true

- name: geam_ex_repeatability_check
  category: stress
  function: geam_ex
  precision: *half_single_double_precisions
  transA_transB: *tiny_transA_transB_range
  matrix_size:
    -  { M:     3, N:    33, K: 15, lda:    35, ldb:    35, ldc:    35, ldd:  85 }
  alpha_beta: *small_alpha_beta_range
  geam_op: [0, 1]
  initialization: hpl
  pointer_mode_host: false
  iters: 5
  atomics_mode : 0
  devices: 4
  repeatability_check: true

...