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

Definitions:

  - &small_matrix_size_range
    - { N:    -1, lda:    -1 }
    - { N:     5, lda:   100 }
    - { N:    15, lda:    16 }
    - { N:    10, lda:    10 }
    - { N:    10, lda:    20 }
    - { N:    20, lda:   160 }
    - { N:    21, lda:    14 }
    - { N:    32, lda:    32 }
    - { N:    48, lda:    48 }
    - { N:    64, lda:    192 }
    - { N:    96, lda:    96 }

  - &medium_matrix_size_range
    - { N:    128, lda:    128 }
    - { N:    224, lda:    224 }
    - { N:    256, lda:    256 }
    - { N:    111, lda:    111 }
    - { N:    160, lda:    160 }
    - { N:    181, lda:    181 }
    - { N:    197, lda:    275 }
    - { N:    352, lda:    352 }
    # below needed to test multiple iterations of remainder loop w/ IB = 32
    - { N:    480, lda:    480 }
    - { N:    500, lda:    500 }
    - { N:    736, lda:    736 }

  - &large_matrix_size_range
    - { N:  357, lda:  400 }
    - { N:  448, lda:  448 }
    - { N:  480, lda:  480 }
    - { N:  511, lda:  512 }
    - { N:  736, lda:  750 }
    - { N:  992, lda: 1000 }
    - { N: 1023, lda: 1023 }
    - { N: 1024, lda: 1024 }
    - { N: 1025, lda: 1025 }
    - { N: 2016, lda: 2016 }

  - &int32_overflow_sizes
    - { N: 8,     lda: 306783380 } # for rocblas_trtri_small
    - { N: 737,   lda: 2921750 } # should take all paths in rocblas_trtri_large
    # - { N: 46976, lda: 46976 } # passed initial tests with increased timeout

Tests:
- name: trtri_bad_arg
  category: quick
  function:
  - trtri_bad_arg: *single_double_precisions_complex_real
  - trtri_batched_bad_arg: *single_double_precisions_complex_real
  - trtri_strided_batched_bad_arg: *single_double_precisions_complex_real
  uplo: L
  diag: N
  api: [ C, FORTRAN ]

- name: trtri_internal
  category: pre_checkin
  function:
    - trtri
    - trtri_batched
    - trtri_strided_batched
  precision: *single_precision
  uplo: [ U, L ]
  diag: [ N ]
  N: [ 9, 736 ]
  lda: [ 2000 ]
  ldc: *c_pos_overflow_int32 # ldc, ldd used for internal api offset args
  ldd: *c_neg_overflow_int32
  stride_scale: [ 1 ]
  batch_count: [ 3 ]
  api: INTERNAL

- name: trtri_quick
  category: quick
  function: trtri
  precision: *single_double_precisions_complex_real
  uplo: [ U, L ]
  diag: [ N, U ]
  matrix_size: *small_matrix_size_range
  api: [ C, FORTRAN ]

- name: trtri_batched_quick
  category: quick
  function:
    - trtri_batched
    - trtri_strided_batched
  precision: *single_double_precisions_complex_real
  uplo: [ U, L ]
  diag: [ N, U ]
  matrix_size: *small_matrix_size_range
  batch_count: [ -1, 0, 1, 25 ]
  api: [ C, FORTRAN ]

- name: trtri_pre
  category: pre_checkin
  function:
    - trtri
    - trtri_batched
    - trtri_strided_batched
  precision: *single_double_precisions_complex_real
  uplo: [ U, L ]
  diag: [ N, U ]
  matrix_size: *medium_matrix_size_range
  batch_count: [ 3 ]

- name: trtri_nightly
  category: nightly
  function:
    - trtri
    - trtri_batched
    - trtri_strided_batched
  precision: *single_double_precisions_complex_real
  uplo: [ U, L ]
  diag: [ N, U ]
  matrix_size: *large_matrix_size_range
  batch_count: [ 2 ]

- name: trtri_large_batch
  category: stress
  arguments:
    - { N: 3, lda: 3, batch_count: *c_grid_yz_require_passes }
  uplo: [ U, L ]
  diag: [ N, U ]
  stride_scale: [ 1 ]
  function:
    - trtri_batched: *single_precision
    - trtri_strided_batched: *single_precision

- name: trtri_size_t
  category: stress_trtri
  function:
    - trtri
    # - trtri_batched
    # - trtri_strided_batched
  precision: *single_double_precisions_complex_real
  uplo: [ U, L ]
  diag: [ N ]
  matrix_size: *int32_overflow_sizes
  gpu_arch: ['90a','942','950']
  os_flags: LINUX

- name: trtri_graph_test
  category: pre_checkin
  function:
    - trtri
    - trtri_batched
    - trtri_strided_batched
  precision: *single_double_precisions_complex_real
  uplo: [ L ]
  diag: [ N ]
  matrix_size:
    - { N:    32, lda:    32 }
  batch_count: [ 2 ]
  graph_test: true


- name: trtri_repeatability_check
  category: stress
  function:
    - trtri
    - trtri_batched
    - trtri_strided_batched
  precision: *single_double_precisions_complex_real
  uplo: [ L ]
  diag: [ N ]
  matrix_size: *medium_matrix_size_range
  batch_count: [ 2 ]
  stride_scale: 1
  atomics_mode: 0
  pointer_mode_host: false
  iters: 5
  initialization: hpl
  devices: 4
  repeatability_check: true
...