derek.holloway
7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
759 lines
25 KiB
YAML
759 lines
25 KiB
YAML
---
|
|
include: rocblas_common.yaml
|
|
include: known_bugs.yaml
|
|
|
|
Definitions:
|
|
- &small_matrix_size_range
|
|
- { M: 10, N: 10, lda: 2, stride_a: 1 }
|
|
- { M: 100, N: 200, lda: 200, stride_a: 40000 }
|
|
|
|
- &sm_mn_matrix_size_range
|
|
# m && n <= 32 && batch_count >= 256 (gfx90a only) watch for overlap with qmcpack
|
|
- { M: 3 , N: 32, lda: 3, stride_a: 96 }
|
|
- { M: 24, N: 24, lda: 24, stride_a: 1024 }
|
|
- { M: 32, N: 11, lda: 32, stride_a: 1024 }
|
|
|
|
- &qmcpack_matrix_size_range
|
|
# m <= 64 && batch_count > 8 (transposes only), N >= 33 to avoid sm_mn kernel
|
|
- { M: 2 , N: 33, lda: 2, stride_a: 66 }
|
|
- { M: 32, N: 34, lda: 32, stride_a: 1088 }
|
|
- { M: 15, N: 35, lda: 16, stride_a: 1024 }
|
|
- { M: 32, N: 1536, lda: 32, stride_a: 49152 }
|
|
|
|
- &skinny_n_matrix_size_range
|
|
# n <= 128 && m >= 2048 * n
|
|
- { M: 32000, N: 11, lda: 32000, stride_a: 352000 }
|
|
- { M: 81920, N: 40, lda: 81920, stride_a: 3276800 }
|
|
- { M: 20480, N: 7, lda: 20480, stride_a: 143360 }
|
|
- { M: 131071, N: 63, lda: 131071, stride_a: 8257473 }
|
|
|
|
- &double_buffered_loads_size_range
|
|
# (n <= 3000 && n >= m) || (n %128 == 0 && m==n)
|
|
- { M: 256, N: 256, lda: 256, stride_a: 65536 }
|
|
- { M: 1024, N: 1024, lda: 1024, stride_a: 1048576 }
|
|
- { M: 3840, N: 3840, lda: 3840, stride_a: 14745600} # test for gfx90a, gfx942, gfx950
|
|
|
|
- &all_algo_matrix_size_range
|
|
- { M: 4096, N: 2, lda: 4096, stride_a: 8192 } # skinny n
|
|
- { M: 32, N: 32, lda: 32, stride_a: 1024 }
|
|
- { M: 100, N: 200, lda: 200, stride_a: 40000 }
|
|
|
|
- &special_case_range
|
|
# Quick return
|
|
- { M: 0, N: 1, lda: 1, incx: 1, incy: 1, batch_count: 1 }
|
|
- { M: 1, N: 0, lda: 1, incx: 1, incy: 1, batch_count: 1 }
|
|
- { M: 1, N: 1, lda: 1, incx: 1, incy: 1, batch_count: 0 }
|
|
|
|
# invalid_arg checks
|
|
- { M: -1, N: 0, lda: 1, incx: 1, incy: 1, batch_count: 0 }
|
|
- { M: 0, N: -1, lda: 1, incx: 1, incy: 1, batch_count: 0 }
|
|
- { M: 5, N: 5, lda: 4, incx: 1, incy: 1, batch_count: 0 }
|
|
- { M: 0, N: 0, lda: 0, incx: 1, incy: 1, batch_count: 0 }
|
|
- { M: 0, N: 0, lda: 1, incx: 0, incy: 1, batch_count: 0 }
|
|
- { M: 0, N: 0, lda: 1, incx: 1, incy: 0, batch_count: 0 }
|
|
- { M: 0, N: 0, lda: 1, incx: 1, incy: 1, batch_count: -1 }
|
|
|
|
- &medium_matrix_size_range
|
|
- { M: 300, N: 400, lda: 400, stride_a: 160000 }
|
|
- { M: 600, N: 500, lda: 601, stride_a: 301000 }
|
|
|
|
- &large_matrix_size_range
|
|
- { M: 1000, N: 1000, lda: 1000, stride_a: 1000000 }
|
|
- { M: 2000, N: 2000, lda: 2000, stride_a: 4000000 }
|
|
- { M: 4011, N: 4011, lda: 4011, stride_a: 16088200 }
|
|
- { M: 8000, N: 8000, lda: 8000, stride_a: 64000000 }
|
|
|
|
- &very_large_matrix_size_range
|
|
- { M: 16010, N: 16010, lda: 16010, stride_a: 256320100 }
|
|
- { M: 25020, N: 25020, lda: 25020, stride_a: 625000400 }
|
|
|
|
- &gemvt_fat_row_vectorized_range
|
|
# Specific sizes to trigger the gemvt_row_vectorized kernels (m << n is required)
|
|
- { M: 16..32..8, N: 512, lda: 32, stride_a: 16384 } # s,d,c,z m<=128, (n >= 8*m on 942)
|
|
- { M: 256, N: 1024, lda: 256, stride_a: 262144 } # d, c, m<=384, (n >= 4*m on 90a)
|
|
- { M: 512, N: 2048, lda: 512, stride_a: 1048576 } # s, m<=768
|
|
- { M: 60, N: 512, lda: 256, stride_a: 131072 } # z, m<=128
|
|
- { M: 300, N: 2400, lda: 300, stride_a: 720000 } # d, c, m<=384
|
|
|
|
- &incx_incy_range
|
|
- { incx: 2, incy: 1, stride_scale: 1 }
|
|
- { incx: -1, incy: 2, stride_scale: 1 }
|
|
- { incx: 1, incy: 1, stride_scale: 1 }
|
|
- { incx: -1, incy: 3, stride_scale: 1.5 }
|
|
- { incx: 3, incy: -1, stride_scale: 2 }
|
|
- { incx: 10, incy: 100, stride_scale: 1 }
|
|
|
|
- &incx_incy_range_small
|
|
- { incx: 2, incy: 2, stride_scale: 1 }
|
|
|
|
- &incx_incy_unity
|
|
- { incx: 1, incy: 1, stride_scale: 1 }
|
|
|
|
- &alpha_beta_range
|
|
- { alpha: 2.0, beta: 0.0, alphai: 1.5, betai: 0.5 }
|
|
- { alpha: -1.0, beta: -1.0, alphai: 0.5, betai: 1.5 }
|
|
- { alpha: 2.0, beta: 1.0, alphai: -1.5, betai: 0.5 }
|
|
- { alpha: 0.0, beta: 1.0, alphai: -0.5, betai: 0 }
|
|
|
|
- &alpha_beta_range_small
|
|
- { alpha: 2.0, beta: 2.0, alphai: 1.5, betai: -1.5 }
|
|
|
|
- &alpha_beta_range_size_t
|
|
- { alpha: 2.0, beta: 2.0, alphai: 1.5, betai: -1.5 }
|
|
- { alpha: 0, beta: 2, alphai: 0, betai: -1.5 }
|
|
- { alpha: 2.0, beta: 0, alphai: 1.5, betai: 0 }
|
|
- { alpha: 0, beta: 0, alphai: 0, betai: 0 }
|
|
|
|
# Non-transpose: batched gfx90a
|
|
# Not a feasable test with current memory contraints
|
|
# - &size_t_large_batch
|
|
# - { M: 2, N: 2, lda: 1073741825, incx: *c_pos_x2_overflow_int32, incy: 1, batch_count: 256 }
|
|
# - { M: 2, N: 2, lda: 2, incx: *c_pos_x2_overflow_int32, incy: 1, batch_count: 256 }
|
|
# - { M: 2, N: 2, lda: 2, incx: *c_neg_x2_overflow_int32, incy: 1, batch_count: 256 }
|
|
# - { M: 2, N: 2, lda: 2, incx: 1, incy: *c_pos_x2_overflow_int32, batch_count: 256 }
|
|
# - { M: 2, N: 2, lda: 2, incx: 1, incy: *c_neg_x2_overflow_int32, batch_count: 256 }
|
|
|
|
# Non-transpose: non-batched 90a tests
|
|
- &size_t_non_transpose_90a
|
|
# skinny
|
|
- { M: 4096, N: 2, lda: 524416, incx: 1, incy: 1 }
|
|
- { M: 4096, N: 2, lda: 4096, incx: 524416, incy: 1 }
|
|
- { M: 4096, N: 2, lda: 4096, incx: -524416, incy: 1 }
|
|
- { M: 4096, N: 2, lda: 4096, incx: 1, incy: 524416 }
|
|
- { M: 4096, N: 2, lda: 4096, incx: 1, incy: -524416 }
|
|
# gfx90a double buffered loads, w/ atomics
|
|
- { M: 128, N: 128, lda: 16909322, incx: 1, incy: 1 }
|
|
- { M: 128, N: 128, lda: 128, incx: 16909322, incy: 1 }
|
|
- { M: 128, N: 128, lda: 128, incx: -16909322, incy: 1 }
|
|
- { M: 128, N: 128, lda: 128, incx: 1, incy: 16909322 }
|
|
- { M: 128, N: 128, lda: 128, incx: 1, incy: -16909322 }
|
|
# general 90a
|
|
- { M: 3, N: 3, lda: *c_pos_x2_overflow_int32, incx: 1, incy: 1 }
|
|
- { M: 3, N: 3, lda: 3, incx: *c_pos_x2_overflow_int32, incy: 1 }
|
|
- { M: 3, N: 3, lda: 3, incx: *c_neg_x2_overflow_int32, incy: 1 }
|
|
- { M: 3, N: 3, lda: 3, incx: 1, incy: *c_pos_x2_overflow_int32 }
|
|
- { M: 3, N: 3, lda: 3, incx: 1, incy: *c_neg_x2_overflow_int32 }
|
|
|
|
# Non-transpose: gfx908 optimization, m < 15000, n < 15000
|
|
- &size_t_908
|
|
- { M: 3, N: 3, lda: *c_pos_x2_overflow_int32, incx: 1, incy: 1 }
|
|
- { M: 3, N: 3, lda: 3, incx: *c_pos_x2_overflow_int32, incy: 1 }
|
|
- { M: 3, N: 3, lda: 3, incx: *c_neg_x2_overflow_int32, incy: 1 }
|
|
- { M: 3, N: 3, lda: 3, incx: 1, incy: *c_pos_x2_overflow_int32 }
|
|
- { M: 3, N: 3, lda: 3, incx: 1, incy: *c_neg_x2_overflow_int32 }
|
|
|
|
# half/bfloat tests: using generic precision kernels currently
|
|
- &size_t_fp16
|
|
- { M: 3, N: 3, lda: *c_pos_x2_overflow_int32, incx: 1, incy: 1, batch_count: 1 }
|
|
- { M: 3, N: 3, lda: 3, incx: *c_pos_x2_overflow_int32, incy: 1, batch_count: 1 }
|
|
- { M: 3, N: 3, lda: 3, incx: *c_neg_x2_overflow_int32, incy: 1, batch_count: 1 }
|
|
- { M: 3, N: 3, lda: 3, incx: 1, incy: *c_pos_x2_overflow_int32, batch_count: 1 }
|
|
- { M: 3, N: 3, lda: 3, incx: 1, incy: *c_neg_x2_overflow_int32, batch_count: 1 }
|
|
|
|
# Transpose/Conj: batched gfx90a
|
|
# Not a feasable test with current memory contraints
|
|
# - &size_t_batch10
|
|
# - { M: 2, N: 2, lda: *c_pos_x2_overflow_int32, incx: 1, incy: 1, batch_count: 9 }
|
|
# - { M: 2, N: 2, lda: 2, incx: *c_pos_x2_overflow_int32, incy: 1, batch_count: 9 }
|
|
# - { M: 2, N: 2, lda: 2, incx: *c_neg_x2_overflow_int32, incy: 1, batch_count: 9 }
|
|
# - { M: 2, N: 2, lda: 2, incx: 1, incy: *c_pos_x2_overflow_int32, batch_count: 9 }
|
|
# - { M: 2, N: 2, lda: 2, incx: 1, incy: *c_neg_x2_overflow_int32, batch_count: 9 }
|
|
|
|
# Transpose/Conj: non-batched 90a tests
|
|
- &size_t_transpose_90a
|
|
# skinny
|
|
- { M: 8192, N: 4, lda: 262177, incx: 1, incy: 1 }
|
|
- { M: 8192, N: 4, lda: 8192, incx: 262177, incy: 1 }
|
|
- { M: 8192, N: 4, lda: 8192, incx: -262177, incy: 1 }
|
|
- { M: 8192, N: 4, lda: 8192, incx: 1, incy: *c_pos_x2_overflow_int32 }
|
|
- { M: 8192, N: 4, lda: 8192, incx: 1, incy: *c_neg_x2_overflow_int32 }
|
|
# general hardware shared-mem reduction
|
|
- { M: 3, N: 3, lda: *c_pos_x2_overflow_int32, incx: 1, incy: 1 }
|
|
- { M: 3, N: 3, lda: 3, incx: *c_pos_x2_overflow_int32, incy: 1 }
|
|
- { M: 3, N: 3, lda: 3, incx: *c_neg_x2_overflow_int32, incy: 1 }
|
|
- { M: 3, N: 3, lda: 3, incx: 1, incy: *c_pos_x2_overflow_int32 }
|
|
- { M: 3, N: 3, lda: 3, incx: 1, incy: *c_neg_x2_overflow_int32 }
|
|
#general case (m & n have to be reasonably large)
|
|
- { M: 4193, N: 4194, lda: 512282, incx: 1, incy: 1 }
|
|
- { M: 4193, N: 4194, lda: 4193, incx: 512282, incy: 1 }
|
|
- { M: 4193, N: 4194, lda: 4193, incx: -512282, incy: 1 }
|
|
- { M: 4193, N: 4194, lda: 4193, incx: 1, incy: 512282 }
|
|
- { M: 4193, N: 4194, lda: 4193, incx: 1, incy: -512282 }
|
|
|
|
# Conj: general case m > 6000, n > 6000 non-float
|
|
- &size_t_general_c_90a
|
|
- { M: 6001, N: 6001, lda: 357914, incx: 1, incy: 1 }
|
|
- { M: 6001, N: 6001, lda: 6001, incx: 357914, incy: 1 }
|
|
- { M: 6001, N: 6001, lda: 6001, incx: -357914, incy: 1 }
|
|
- { M: 6001, N: 6001, lda: 6001, incx: 1, incy: 357914 }
|
|
- { M: 6001, N: 6001, lda: 6001, incx: 1, incy: -357914 }
|
|
|
|
# Transpose/Conj: 908 double buffered loads, atomics allowed
|
|
- &size_t_db_loads_t_908
|
|
- { M: 10496, N: 10496, lda: 204620, incx: 1, incy: 1 }
|
|
- { M: 10496, N: 10496, lda: 10496, incx: 204620, incy: 1 }
|
|
- { M: 10496, N: 10496, lda: 10496, incx: 204620, incy: 1 }
|
|
- { M: 10496, N: 10496, lda: 10496, incx: 1, incy: 204620 }
|
|
- { M: 10496, N: 10496, lda: 10496, incx: 1, incy: 204620 }
|
|
|
|
# Transpose: gfx10/11/12 warp-reduction
|
|
- &size_t_gfx10_11_12_wr
|
|
- { M: 3, N: 3, lda: *c_pos_x2_overflow_int32, incx: 1, incy: 1 }
|
|
- { M: 3, N: 3, lda: 3, incx: *c_pos_x2_overflow_int32, incy: 1 }
|
|
- { M: 3, N: 3, lda: 3, incx: *c_neg_x2_overflow_int32, incy: 1 }
|
|
- { M: 3, N: 3, lda: 3, incx: 1, incy: *c_pos_x2_overflow_int32 }
|
|
- { M: 3, N: 3, lda: 3, incx: 1, incy: *c_neg_x2_overflow_int32 }
|
|
|
|
Tests:
|
|
# Regular gemv
|
|
- name: gemv_bad_arg
|
|
category: quick
|
|
function: gemv_bad_arg
|
|
precision: *single_double_precisions_complex_real
|
|
transA: N
|
|
api: [ FORTRAN, FORTRAN_64 ]
|
|
|
|
- name: gemv_arg_check
|
|
category: quick
|
|
function: gemv
|
|
precision: *single_double_precisions
|
|
transA: N
|
|
|
|
- name: gemv_NaN
|
|
category: pre_checkin
|
|
function: gemv
|
|
precision: *single_double_precisions
|
|
transA: [ N, T ]
|
|
matrix_size: *all_algo_matrix_size_range
|
|
incx_incy: *incx_incy_range_small
|
|
alpha: [ 1.0, .NaN ] # NaN is converted to 0.0 in test code
|
|
beta: [ 0.5, 1.0, .NaN ]
|
|
|
|
- name: gemv_fortran
|
|
category: quick
|
|
function: gemv
|
|
precision: *single_double_precisions_complex_real
|
|
transA: [ N, T, C ]
|
|
incx_incy: *incx_incy_range_small
|
|
alpha_beta: *alpha_beta_range_small
|
|
api: FORTRAN
|
|
|
|
- name: gemv_small
|
|
category: quick
|
|
function: gemv
|
|
precision: *single_double_precisions_complex_real
|
|
transA: [ N, T, C ]
|
|
matrix_size: *small_matrix_size_range
|
|
incx_incy: *incx_incy_range
|
|
alpha_beta: *alpha_beta_range
|
|
|
|
- name: gemv_medium
|
|
category: pre_checkin
|
|
function: gemv
|
|
precision: *single_double_precisions_complex_real
|
|
transA: [ N, T, C ]
|
|
matrix_size: *medium_matrix_size_range
|
|
incx_incy: *incx_incy_range
|
|
alpha_beta: *alpha_beta_range
|
|
|
|
- name: gemv_double_buffered
|
|
category: pre_checkin
|
|
function: gemv
|
|
precision: *single_double_precisions
|
|
transA: [ N ]
|
|
matrix_size: *double_buffered_loads_size_range
|
|
incx_incy: *incx_incy_range_small
|
|
alpha_beta: *alpha_beta_range_small
|
|
atomics_mode: [0, 1]
|
|
|
|
- name: gemv_medium_HMM
|
|
category: HMM
|
|
function: gemv
|
|
precision: *single_double_precisions_complex_real
|
|
transA: [ N ]
|
|
matrix_size: *medium_matrix_size_range
|
|
incx: -1
|
|
incy: 2
|
|
alpha: 1
|
|
beta: 1
|
|
|
|
- name: gemv_medium_multi_gpu
|
|
category: multi_gpu
|
|
function: gemv
|
|
precision: *single_double_precisions
|
|
transA: [ N, T ]
|
|
matrix_size: *medium_matrix_size_range
|
|
incx: 2
|
|
incy: 1
|
|
alpha_beta: *alpha_beta_range
|
|
threads_streams: *common_threads_streams
|
|
devices: [0, 2, 4]
|
|
|
|
- name: gemv_medium_multi_threads_streams
|
|
category: nightly
|
|
function: gemv
|
|
precision: *single_double_precisions_complex_real
|
|
transA: [ N, T, C ]
|
|
matrix_size: *medium_matrix_size_range
|
|
incx: 2
|
|
incy: 1
|
|
alpha_beta: *alpha_beta_range
|
|
threads_streams: *common_threads_streams
|
|
|
|
- name: gemv_large
|
|
category: nightly
|
|
function: gemv
|
|
precision: *single_double_precisions_complex_real
|
|
transA: [ N, T, C ]
|
|
matrix_size: *large_matrix_size_range
|
|
incx_incy: *incx_incy_range_small
|
|
alpha_beta: *alpha_beta_range_small
|
|
|
|
- name: gemv_very_large
|
|
category: nightly
|
|
function: gemv
|
|
precision: *double_precision
|
|
transA: [ N ]
|
|
matrix_size: *very_large_matrix_size_range
|
|
incx_incy: *incx_incy_range_small
|
|
alpha_beta: *alpha_beta_range_small
|
|
gpu_arch: '9??'
|
|
|
|
- name: gemv_skinny_n
|
|
category: pre_checkin
|
|
function: gemv
|
|
precision: *single_double_precisions_complex_real
|
|
transA: [ N, T, C ]
|
|
matrix_size: *skinny_n_matrix_size_range
|
|
incx_incy: *incx_incy_range_small
|
|
alpha_beta: *alpha_beta_range
|
|
|
|
- name: gemv_size_t_non_transpose_90a
|
|
category: stress
|
|
function: gemv
|
|
precision: *single_precision
|
|
transA: [ N ]
|
|
matrix_size: *size_t_non_transpose_90a
|
|
alpha_beta: *alpha_beta_range_size_t
|
|
pointer_mode_host: false
|
|
atomics_mode: [0, 1]
|
|
os_flags: LINUX
|
|
gpu_arch: ['90a']
|
|
|
|
- name: gemv_size_t_non_transpose_908
|
|
category: stress
|
|
function: gemv
|
|
precision: *single_precision
|
|
transA: [ N ]
|
|
matrix_size: *size_t_908
|
|
alpha_beta: *alpha_beta_range_size_t
|
|
pointer_mode_host: false
|
|
os_flags: LINUX
|
|
gpu_arch: ['908']
|
|
|
|
- name: gemv_size_t_non_transpose_fp16_90a
|
|
category: stress
|
|
function: gemv_batched
|
|
precision: *hpa_half_precision
|
|
transA: [ N ]
|
|
matrix_size: *size_t_fp16
|
|
alpha_beta: *alpha_beta_range_size_t
|
|
pointer_mode_host: false
|
|
os_flags: LINUX
|
|
gpu_arch: ['90a']
|
|
|
|
- name: gemv_size_t_transpose_90a
|
|
category: stress
|
|
function: gemv
|
|
precision: *single_precision
|
|
transA: [ T, C ]
|
|
matrix_size: *size_t_transpose_90a
|
|
alpha_beta: *alpha_beta_range_size_t
|
|
pointer_mode_host: false
|
|
os_flags: LINUX
|
|
gpu_arch: ['90a']
|
|
|
|
- name: gemv_size_t_conj_90a
|
|
category: stress
|
|
function: gemv
|
|
precision: *double_precision
|
|
transA: [ T, C ]
|
|
matrix_size: *size_t_general_c_90a
|
|
alpha_beta: *alpha_beta_range_size_t
|
|
pointer_mode_host: false
|
|
os_flags: LINUX
|
|
gpu_arch: ['90a']
|
|
|
|
- name: gemv_size_t_transpose_908
|
|
category: stress
|
|
function: gemv
|
|
precision: *single_precision
|
|
transA: [ T, C ]
|
|
matrix_size: *size_t_db_loads_t_908
|
|
alpha_beta: *alpha_beta_range_size_t
|
|
pointer_mode_host: false
|
|
os_flags: LINUX
|
|
gpu_arch: ['908']
|
|
|
|
- name: gemv_size_t_transpose_10_11_12
|
|
category: stress
|
|
function: gemv
|
|
precision: *single_precision
|
|
transA: [ T, C ]
|
|
matrix_size: *size_t_gfx10_11_12_wr
|
|
alpha_beta: *alpha_beta_range_size_t
|
|
pointer_mode_host: false
|
|
os_flags: LINUX
|
|
gpu_arch: ['10??', '11??', '12??']
|
|
|
|
# gemv_batched
|
|
- name: gemv_batched_bad_arg
|
|
category: quick
|
|
function: gemv_batched_bad_arg
|
|
precision: *gemv_bfloat_half_single_double_complex_real_precisions
|
|
transA: N
|
|
api: [ FORTRAN, FORTRAN_64 ]
|
|
|
|
- name: gemv_batched_arg_check
|
|
category: quick
|
|
function: gemv_batched
|
|
precision: *gemv_bfloat_half_single_double_complex_real_precisions
|
|
transA: N
|
|
|
|
- name: gemv_batched_NaN
|
|
category: pre_checkin
|
|
function: gemv_batched
|
|
precision: *gemv_bfloat_half_single_double_complex_real_precisions
|
|
transA: [ N, T, C ]
|
|
matrix_size: *all_algo_matrix_size_range
|
|
incx_incy: *incx_incy_range_small
|
|
alpha: [ 1.0, 2.0, .NaN ] # NaN is converted to 0.0 in test code
|
|
beta: [2.0, 1.0, .NaN ]
|
|
batch_count: [ 2, 9 ] # 9 for qmc_pack algo
|
|
|
|
- name: gemv_batched_fortran
|
|
category: quick
|
|
function: gemv_batched
|
|
precision: *gemv_bfloat_half_single_double_complex_real_precisions
|
|
transA: [ N, T, C ]
|
|
incx_incy: *incx_incy_range_small
|
|
alpha_beta: *alpha_beta_range_small
|
|
batch_count: [ -1, 0, 3 ]
|
|
api: FORTRAN
|
|
|
|
- name: gemv_batched_double_buffered
|
|
category: pre_checkin
|
|
function: gemv_batched
|
|
precision: *single_double_precisions
|
|
transA: [ N ]
|
|
matrix_size: *double_buffered_loads_size_range
|
|
incx_incy: *incx_incy_range_small
|
|
alpha_beta: *alpha_beta_range_small
|
|
batch_count: [ 3 ]
|
|
atomics_mode: [0, 1]
|
|
|
|
- name: gemv_batched_skinny_n
|
|
category: pre_checkin
|
|
function: gemv_batched
|
|
precision: *single_double_precisions_complex_real
|
|
transA: [ N, T, C ]
|
|
matrix_size: *skinny_n_matrix_size_range
|
|
incx_incy: *incx_incy_range_small
|
|
alpha_beta: *alpha_beta_range_small
|
|
batch_count: [ 3 ]
|
|
|
|
- name: gemv_batched_small
|
|
category: quick
|
|
function: gemv_batched
|
|
precision: *single_double_precisions
|
|
transA: [ N, T, C ]
|
|
matrix_size: *small_matrix_size_range
|
|
incx_incy: *incx_incy_range
|
|
alpha_beta: *alpha_beta_range
|
|
batch_count: [ -1, 0, 1, 3 ]
|
|
|
|
- name: gemv_sm_mn_batched
|
|
category: pre_checkin
|
|
function:
|
|
- gemv_batched
|
|
- gemv_strided_batched
|
|
precision: *gemv_bfloat_half_single_double_complex_real_precisions
|
|
transA: [ N ]
|
|
matrix_size: *sm_mn_matrix_size_range
|
|
incx_incy: *incx_incy_range_small
|
|
alpha: [ 1.0, .NaN ] # NaN is converted to 0.0 in test code
|
|
beta: [ 1.0, .NaN ]
|
|
batch_count: [256, 513] # >= 256
|
|
|
|
- name: gemv_batched_qmcpack
|
|
category: quick
|
|
function: gemv_batched
|
|
precision: *single_double_precisions_complex_real
|
|
transA: [ T, C ]
|
|
matrix_size: *qmcpack_matrix_size_range
|
|
incx: [ 2 ]
|
|
incy: [ 2 ]
|
|
alpha_beta: *alpha_beta_range
|
|
batch_count: [ 100, 1000 ]
|
|
|
|
- name: gemv_batched_medium
|
|
category: pre_checkin
|
|
function: gemv_batched
|
|
precision: *single_double_precisions_complex_real
|
|
transA: [ N, T, C ]
|
|
matrix_size: *medium_matrix_size_range
|
|
incx_incy: *incx_incy_range
|
|
alpha_beta: *alpha_beta_range
|
|
batch_count: [ 3 ]
|
|
|
|
- name: gemv_batched_large
|
|
category: nightly
|
|
function: gemv_batched
|
|
precision: *gemv_bfloat_half_single_double_complex_real_precisions
|
|
transA: [ N, T, C ]
|
|
matrix_size: *large_matrix_size_range
|
|
incx_incy: *incx_incy_range_small
|
|
alpha_beta: *alpha_beta_range_small
|
|
batch_count: [ 3 ]
|
|
|
|
- name: gemv_batched_very_large
|
|
category: nightly
|
|
function: gemv_batched
|
|
precision: *double_precision
|
|
transA: [ T ]
|
|
matrix_size: *very_large_matrix_size_range
|
|
incx_incy: *incx_incy_range_small
|
|
alpha_beta: *alpha_beta_range_small
|
|
batch_count: [ 2 ]
|
|
gpu_arch: '9??'
|
|
|
|
- name: gemv_batched_grid_z_16bit
|
|
category: stress
|
|
arguments:
|
|
- { M: 2048, N: 1, lda: 2048, incx: 1, incy: 1, batch_count: *c_grid_yz_require_passes} #skinny tuned block size gemvn_kernel
|
|
- { M: 128, N: 128, lda: 128, incx: 1, incy: 1, batch_count: *c_grid_yz_require_passes} #double buffered loads
|
|
- { M: 3, N: 3, lda: 3, incx: 1, incy: 1, batch_count: *c_grid_yz_require_passes} #generic kernel
|
|
transA: [ N, T, C ]
|
|
alpha_beta: *alpha_beta_range_small
|
|
pointer_mode_host: false
|
|
function:
|
|
- gemv_batched: *single_precision
|
|
- gemv_strided_batched: *single_precision
|
|
stride_scale: 1
|
|
|
|
# gemv_strided_batched
|
|
- name: gemv_strided_batched_bad_arg
|
|
category: quick
|
|
function: gemv_strided_batched_bad_arg
|
|
precision: *gemv_bfloat_half_single_double_complex_real_precisions
|
|
transA: N
|
|
api: [ FORTRAN, FORTRAN_64 ]
|
|
|
|
- name: gemv_strided_batched_arg_check
|
|
category: quick
|
|
function: gemv_strided_batched
|
|
precision: *gemv_bfloat_half_single_double_complex_real_precisions
|
|
transA: N
|
|
|
|
- name: gemv_strided_batched_NaN
|
|
category: pre_checkin
|
|
function: gemv_strided_batched
|
|
precision: *gemv_bfloat_half_single_double_complex_real_precisions
|
|
transA: [ N, T, C ]
|
|
matrix_size: *all_algo_matrix_size_range
|
|
incx_incy: *incx_incy_range_small
|
|
alpha: [ 2.0, .NaN ] # NaN converted to 0.0 in test code
|
|
beta: [ 0.5, 1.0, .NaN ]
|
|
batch_count: [ 2 ]
|
|
|
|
- name: gemv_strided_batched_fortran
|
|
category: quick
|
|
function: gemv_strided_batched
|
|
precision: *single_double_precisions_complex_real
|
|
transA: [ N, T, C ]
|
|
incx_incy: *incx_incy_range_small
|
|
alpha_beta: *alpha_beta_range_small
|
|
batch_count: [ -1, 0, 3 ]
|
|
api: FORTRAN
|
|
|
|
- name: gemv_strided_batched_small
|
|
category: quick
|
|
function: gemv_strided_batched
|
|
precision: *gemv_bfloat_half_single_double_complex_real_precisions
|
|
transA: [ N, T, C ]
|
|
matrix_size: *small_matrix_size_range
|
|
incx_incy: *incx_incy_range
|
|
alpha_beta: *alpha_beta_range
|
|
batch_count: [ -1, 0, 1, 3 ]
|
|
|
|
- name: gemv_strided_batched_double_buffered
|
|
category: pre_checkin
|
|
function: gemv_strided_batched
|
|
precision: *single_double_precisions
|
|
transA: [ N ]
|
|
matrix_size: *double_buffered_loads_size_range
|
|
incx_incy: *incx_incy_range_small
|
|
alpha_beta: *alpha_beta_range_small
|
|
batch_count: [ 3 ]
|
|
atomics_mode: [0, 1]
|
|
|
|
- name: gemv_strided_batched_skinny_n
|
|
category: pre_checkin
|
|
function: gemv_strided_batched
|
|
precision: *gemv_bfloat_half_single_double_complex_real_precisions
|
|
transA: [ N, T, C ]
|
|
matrix_size: *skinny_n_matrix_size_range
|
|
incx_incy: *incx_incy_range_small
|
|
alpha_beta: *alpha_beta_range_small
|
|
batch_count: [ 3 ]
|
|
|
|
|
|
- name: gemv_strided_batched_qmcpack
|
|
category: quick
|
|
function: gemv_strided_batched
|
|
precision: *gemv_bfloat_half_single_double_complex_real_precisions
|
|
transA: [ T, C ]
|
|
matrix_size: *qmcpack_matrix_size_range
|
|
incx: [ 2 ]
|
|
incy: [ 2 ]
|
|
stride_scale: 1
|
|
alpha_beta: *alpha_beta_range
|
|
batch_count: [ 64, 512 ]
|
|
|
|
- name: gemv_strided_batched_medium
|
|
category: pre_checkin
|
|
function: gemv_strided_batched
|
|
precision: *gemv_bfloat_half_single_double_complex_real_precisions
|
|
transA: [ N, T, C ]
|
|
matrix_size: *medium_matrix_size_range
|
|
incx_incy: *incx_incy_range
|
|
alpha_beta: *alpha_beta_range
|
|
batch_count: [ 3 ]
|
|
|
|
- name: gemv_strided_batched_large
|
|
category: nightly
|
|
function: gemv_strided_batched
|
|
precision: *single_double_precisions
|
|
transA: [ N, T, C ]
|
|
matrix_size: *large_matrix_size_range
|
|
incx_incy: *incx_incy_range_small
|
|
alpha_beta: *alpha_beta_range_small
|
|
batch_count: [ 3 ]
|
|
|
|
- name: gemv_strided_batched_very_large
|
|
category: nightly
|
|
function: gemv_strided_batched
|
|
precision: *gemv_bfloat_half_single_double_complex_real_precisions
|
|
transA: [ N, T ]
|
|
matrix_size: *very_large_matrix_size_range
|
|
incx_incy: *incx_incy_range_small
|
|
alpha_beta: *alpha_beta_range_small
|
|
batch_count: [ 2 ]
|
|
gpu_arch: '9??'
|
|
|
|
- name: gemv_stride_alpha_beta
|
|
category: pre_checkin
|
|
function:
|
|
- gemv_batched
|
|
- gemv_strided_batched
|
|
precision: *gemv_bfloat_half_single_double_complex_real_precisions
|
|
transA: [ N, T, C ]
|
|
matrix_size: *qmcpack_matrix_size_range
|
|
incx_incy: *incx_incy_range
|
|
alpha_beta: *alpha_beta_range
|
|
batch_count: [ 3 ]
|
|
alpha_beta_stride: true
|
|
stride_c: 2 # alpha beta stride
|
|
stride_d: 2
|
|
pointer_mode_host: false
|
|
|
|
- name: gemv_row_vectorized_coverage
|
|
category: pre_checkin
|
|
function:
|
|
- gemv
|
|
- gemv_batched
|
|
- gemv_strided_batched
|
|
precision: *single_double_precisions_complex_real
|
|
transA: [ T, C ]
|
|
matrix_size: *gemvt_fat_row_vectorized_range
|
|
incx_incy: *incx_incy_range
|
|
alpha_beta: *alpha_beta_range
|
|
batch_count: [ 33, 256 ]
|
|
gpu_arch: ['90a', '942']
|
|
|
|
- name: gemv_ILP64
|
|
category: stress
|
|
arguments:
|
|
- { M: 2, N: 2, lda: 2147483649, incx: 1, incy: 1, batch_count: 1 }
|
|
- { M: 2, N: 2, lda: 2, incx: 2147483649, incy: 1, batch_count: 1 }
|
|
- { M: 2, N: 2, lda: 2, incx: 1, incy: 2147483649, batch_count: 1 }
|
|
- { M: 2, N: 2, lda: 2, stride_a: 4, incx: 1, incy: 1, batch_count: *c_grid_yz_require_passes }
|
|
transA: [ N, T, C ]
|
|
alpha_beta: *alpha_beta_range_small
|
|
pointer_mode_host: false
|
|
api: [ C_64 ]
|
|
os_flags: [ LINUX ]
|
|
function:
|
|
- gemv: *single_precision
|
|
- gemv_batched: *hpa_half_precision
|
|
- gemv_strided_batched: *hpa_half_precision
|
|
stride_scale: 1
|
|
gpu_arch: '9??'
|
|
|
|
- name: gemv_ILP64
|
|
category: stress
|
|
arguments:
|
|
- { transA: N, M: 2147483649, N: 1, lda: 2147483649, incx: 1, incy: 1, batch_count: 1 }
|
|
- { transA: N, M: 1, N: 2147483649, lda: 1, incx: 1, incy: 1, batch_count: 1 }
|
|
- { transA: T, M: 2147483649, N: 1, lda: 2147483649, incx: 1, incy: 1, batch_count: 1 }
|
|
- { transA: T, M: 1, N: 2147483649, lda: 1, incx: 1, incy: 1, batch_count: 1 }
|
|
alpha_beta: *alpha_beta_range_small
|
|
pointer_mode_host: false
|
|
initialization: hpl # large reductions so ints can overflow
|
|
api: [ C_64 ]
|
|
os_flags: [ LINUX ]
|
|
function:
|
|
- gemv: *single_precision
|
|
# Too slow for now
|
|
#- gemv_batched: *hpa_half_precision
|
|
#- gemv_strided_batched: *hpa_half_precision
|
|
stride_scale: 1
|
|
gpu_arch: '9??'
|
|
|
|
- name: gemv_graph_test
|
|
category: pre_checkin
|
|
function:
|
|
- gemv
|
|
- gemv_batched
|
|
- gemv_strided_batched
|
|
precision: *gemv_bfloat_half_single_double_complex_real_precisions
|
|
transA: N
|
|
matrix_size:
|
|
- { M: 100, N: 200, lda: 200, stride_a: 40000 }
|
|
incx_incy: *incx_incy_range_small
|
|
alpha_beta: *alpha_beta_range_small
|
|
batch_count: [ 5 ]
|
|
graph_test: true
|
|
|
|
- name: gemv_repeatability_check
|
|
category: stress
|
|
function:
|
|
- gemv
|
|
- gemv_batched
|
|
- gemv_strided_batched
|
|
precision: *single_double_precisions
|
|
initialization: hpl
|
|
transA: [ N, T]
|
|
matrix_size:
|
|
- { M: 10496, N: 10496, lda: 10496, incx: 1, incy: 1 }
|
|
alpha: [ 1.0 ]
|
|
atomics_mode: 0
|
|
batch_count: [ 2 ]
|
|
iters: 5
|
|
stride_scale: 1
|
|
pointer_mode_host: false
|
|
devices: 4
|
|
repeatability_check: true
|
|
|
|
...
|