Граф коммитов

191 Коммитов

Автор SHA1 Сообщение Дата
Captain Jack Sparrow faa1e5c618
Merge pull request #101 from microsoft/dev/ritdas/fix_args
Revert some of the changes from the previous commit regarding input_set args
2023-05-09 09:08:05 -07:00
Ritwik Das 93335be767 Revert some of the changes from the previous commit regarding input_set args 2023-05-09 08:43:52 -07:00
Captain Jack Sparrow b5145a4b8e
Merge pull request #100 from microsoft/dev/ritdas/fix_strides
Fix strides of randomly generated data
2023-05-08 08:15:27 -07:00
Ritwik Das d12a4cb7b3 Another fix 2023-05-08 08:06:34 -07:00
Ritwik Das 8ed31c8079 Some cleanup 2023-05-08 07:56:15 -07:00
Ritwik Das b02b922559 Another fix 2023-05-08 06:33:48 -07:00
Ritwik Das 2e2ad05b6e Fix strides of randomly generated data 2023-05-08 06:19:26 -07:00
Captain Jack Sparrow 00e7471dc2
Merge pull request #99 from microsoft/dev/ritdas/verbose_comment
Hide info message under the verbose flag and refactor batch runner for benchmarks
2023-04-12 09:25:09 +05:30
Ritwik Das e3d0834be1 PR comments 2023-04-11 20:38:17 -07:00
Ritwik Das 2df22de71b Refactor batch runner code 2023-04-11 14:10:11 -07:00
Ritwik Das 05880c3b05 Add helper 2023-04-11 12:55:17 -07:00
Ritwik Das d5313bee6f Hide info message under the verbose flag 2023-04-11 12:29:10 -07:00
Kern Handa 7af2531403
Fixes call to __call__ method for CallableFunc subclasses (#98)
Co-authored-by: Lisa Ong <onglisa@microsoft.com>
2023-03-15 16:28:23 -07:00
Lisa Ong 9cefcd7b6f
[Cleanup] Remove/update tests that randomly fail (#97)
* Update add_partial_dynamic to ensure input dimensions are within range
* Replace Range with a simpler function to test runtime arrays
* Re-ran tests 4 times without failures (fingers crossed)
2023-03-15 08:55:23 +08:00
Lisa Ong 2691a987ef
benchmark_hat_package: --working_dir option and resolving `printf` symbols (#96)
* Adds a `--working_dir` option to hatlib.benchmark_hat_package to bypass "path length exceeded" errors on Windows. Defaults to the current working directory.

* [Windows] Temporarily resolve the `printf` symbol for compiled hat packages that do not explicitly include stdio.h. This is temporary and should be declared in the hat package. For example, Accera's region profiling injects printf calls that need to be resolved at link time.

* Removed noisy/flaky test
2023-03-14 09:29:22 +08:00
Kern Handa 620aecc577
Fixes logic in wCUDA and ROCM loader (#95) 2023-03-03 00:56:45 -08:00
Kern Handa 72142154b7
[benchmark] Results will now be appended to the results file (#94) 2023-03-02 23:51:54 -08:00
Kern Handa 18bc07af7f
Fixes numerous issues for benchmarking (#93)
* Fixes numerous issues for benchmarking

* Batch size was used in place of number of iterations per batch size when averaging batch size results.
* C++ benchmarking was developed on top of CallableFunc, which was designed for devices where multiple input sets were not needed. C++ benchmarking on the CPU, at least, needs these multiple input sets. This change fixes this issue.
* Number of iterations per batch was added as a command line argument
* CallableFunc instances for devices that need cache flushing will have extra data generated for them
2023-03-02 19:28:49 -08:00
Kern Handa 71af80bb39
Adds CMakeLists.txt.in to final built package (#92) 2023-03-01 15:18:41 -08:00
Captain Jack Sparrow 758e636ced
Merge pull request #91 from microsoft/dev/ritdas/mintime0
Allow benchmark to run once when min_time_in_sec == 0
2023-02-28 02:35:52 -08:00
Ritwik Das 193c6a329d Allow benchmark to run once when min_time_in_sec == 0 2023-02-28 15:40:14 +05:30
Captain Jack Sparrow ee298d6a16
Merge pull request #89 from microsoft/dev/ritdas/cleanup_fix
Fix cleanup logic for Windows during native profiling
2023-02-15 16:28:46 -08:00
Ritwik Das ad0c03ed5d Fix cleanup logic for Windows during native profiling 2023-02-15 13:38:09 -08:00
Captain Jack Sparrow 1ce9d77ce3
Merge pull request #88 from microsoft/dev/ritdas/path_fix
Fix path to static lib for native profiling
2023-02-14 17:04:10 -08:00
Ritwik Das 22d1672e0b Fix path to static lib for native profiling 2023-02-14 16:36:39 -08:00
Captain Jack Sparrow ff279692bb
Merge pull request #87 from microsoft/dev/ritdas/native_profiling
Support native profiling in benchmark_hat_package
2023-02-13 10:42:48 -08:00
Ritwik Das 6027b22770 PR comments and Role name fix 2023-02-13 10:35:15 -08:00
Ritwik Das 238085a124 Some more changes 2023-02-11 13:19:27 -08:00
Ritwik Das f2bbef790d try fix 2023-02-11 13:13:13 -08:00
Ritwik Das 1e2f1cd861 Undo a change 2023-02-10 23:58:33 -08:00
Ritwik Das 519f53336a Undo a change 2023-02-10 23:52:26 -08:00
Ritwik Das 5478588794 Support native profiling in benchmark_hat_package 2023-02-10 19:24:51 -08:00
Lisa Ong 91e62367cb
Minimal support for caller-allocated dynamic-shaped hat functions (#86)
* minimal special case for two_pass_alloc

* flag TODO

* minimal two-pass alloc support

* fix the stride computation, now that runtime arrays don't have it specified

* PR feedback

* WIP: workaround flaky tests

* add support for bool element type

* add a get_function(name) utility

* maybe fix crashing Range test

Co-authored-by: Lisa Ong <onglisa@microsoft.com>
2023-01-16 17:52:41 +08:00
Denny Sun 8dbb06bace
Add support for floating type arguments to hat functions (#85)
* add array size as arguments

* fix ctypes conversion

* Support more integer types for dynamic dimension

* Update the way of checking integer type

* Support floating type args

* accera-gpu is needed for running the tests
2023-01-06 01:49:13 -08:00
Kern Handa 1f59252ae1
Generates random data for constant shaped args (#84) 2022-12-01 00:49:29 -08:00
Kern Handa 6d93d2955d
Updates RNG for int types (#82) 2022-11-29 21:22:49 -08:00
Kern Handa ab36ccf117
Adds input data process fn hook for benchmarking (#83) 2022-11-29 20:32:29 -08:00
Kern Handa 262252f36f
Fixes more issues in dynamic shape benchmarking (#81) 2022-11-15 17:52:10 -08:00
Kern Handa f90ac1fb86
Fixes dynamic shaped benchmark size usage (#80)
Fixes dynamic shaped benchmark size usage

This change sets the symbolic dimension values using the shapes
provided by the dynamic shape callback function.

It also fixes incorrectly setting the strides in ArgInfo.
2022-11-03 17:54:29 -07:00
Lisa Ong 5aa080c110
Fixes for mixture of output and input_output parameters (#79)
* parameter value incorrectly set for affine array

* handle shape=() ndarrays in args verification

* test fix: order is not guaranteed(?)

* actually drop hash suffix

* .

* flag possible bug

* add helper to lookup function by name

* undo

Co-authored-by: Lisa Ong <onglisa@microsoft.com>
2022-11-01 10:29:08 +08:00
Kern Handa 48b711cc77
Adds the ability to specify a callback fn to specify dynamic shapes for benchmarking (#78) 2022-10-25 05:24:26 -07:00
Kern Handa 610047e571
Updates HATPackage with addition of benchmak and func iterator (#77) 2022-10-10 11:51:52 -07:00
Denny Sun d95d3e64bf
Support all integer types for dynamic dimensions (#76)
* add array size as arguments

* fix ctypes conversion

* Support more integer types for dynamic dimension

* Update the way of checking integer type
2022-10-04 19:47:42 -07:00
Captain Jack Sparrow 9932df4f5d
Merge pull request #75 from microsoft/dev/ritdas/dynamic_max_mem
Set CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES attribute for dynamic shared memory allocation
2022-10-04 15:56:34 -07:00
Captain Jack Sparrow 743e81d1ac Set CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES attribute for dynamic shared memory allocation 2022-10-04 15:08:37 -07:00
Kern Handa 8d4c04632b
Updates ArgValue to better handle constant sizes for runtime arrays (#72) 2022-10-04 12:57:54 -07:00
Captain Jack Sparrow 7dac740c08
Merge pull request #74 from microsoft/dev/ritdas/dynamic_sharedmem
HAT support for dynamic shared memory allocation for GPU
2022-10-03 16:20:10 -07:00
Ritwik Das ddbcd98896 HAT support for dynamic shared memory allocation for GPU 2022-10-03 10:55:01 -07:00
Kern Handa 12e9f2f7f8
Updates test requirements.txt (#73) 2022-09-28 14:14:26 -07:00
Lisa Ong 44badf7705
Function calls with runtime output arrays (#71)
* support partial dynamic sized runtime arrays

* marked TODO

* layout TODO

* fix

* expand arguments to include dimensions automatically

* remove accidental import

* working on correctness

* support calling dynamic sized functions

* return output arguments automatically

* add multi-output test

* nfc

* yapf

* allow output arrays to be passed as input args

* nfc

* support sharing of output dim args

* add check

* fix

* PR feedback

Co-authored-by: Lisa Ong <onglisa@microsoft.com>
2022-09-27 08:31:49 +08:00