Teach Python TS frontend to parse complex literals #52881

anjali411 · 2021-02-25T22:27:54Z

Stack from ghstack:

Add support for real and imag tensor attributes #54542 Add support for real and imag tensor attributes
[TS] Add complex support for more ops #54541 [TS] Add complex support for more ops
Allow creating SugaredValue for a complex valued IValue and deserialization logic for "infj" and "nanj" global constants #54328 Allow creating SugaredValue for a complex valued IValue and deserialization logic for "infj" and "nanj" global constants
Add JIT support for cmath unary ops #54089 Add JIT support for cmath unary ops
Teach Python TS frontend to parse complex literals #52881 Teach Python TS frontend to parse complex literals

This PR adds:

logic to parse complex constants (complex literals of the form bj)
logic to parse complex lists
support for complex constructors: complex(tensor/int/float/bool, tensor/int/float/bool)
Limited operator support
- add, sub, mul, torch.tensor, torch.as_tensor

Follow-up work:

Add complex support for unary and other registered ops.
support complex constructor with string as input (this is supported in Python eager mode).
Test all emitXYZ for all XYZ in ir_emitter.cpp (currently only emitConst, emitValueToTensor are tested). e.g., test loops etc.
onnx doesn't support complex tensors, so we should error out with a clear and descriptive error message.

Differential Revision: D27245059

[ghstack-poisoned]

facebook-github-bot · 2021-02-25T22:28:13Z

💊 CI failures summary and remediations

As of commit e851c1d (more details on the Dr. CI page):

2/3 failures possibly* introduced in this PR
- 1/2 non-scanned failure(s)
1/3 broken upstream at merge base 91d37d7 since Mar 22

🕵️ 1 new failure recognized by patterns

The following CI failures do not appear to be due to upstream breakages:

pytorch_xla_linux_bionic_py3_6_clang9_test (1/1)

Step: "Run tests" (full log | diagnosis details | 🔁 rerun)

Mar 24 03:01:27 ERROR: test_trace_and_metrics (__main__.ProfilerTest)

Mar 24 03:01:27 No trace event is collected. Automatically retrying.
Mar 24 03:01:27 
Mar 24 03:01:27 Starting to trace for 5000 ms. Remaining attempt(s): 0
Mar 24 03:01:27 2021-03-24 03:01:27.575034: E tensorflow/core/profiler/rpc/client/profiler_client.cc:154] Unavailable: failed to connect to all addresses
Mar 24 03:01:27 2021-03-24 03:01:27.575072: W tensorflow/core/profiler/rpc/client/capture_profile.cc:133] No trace event is collected from localhost:59597
Mar 24 03:01:27 2021-03-24 03:01:27.575087: W tensorflow/core/profiler/rpc/client/capture_profile.cc:145] localhost:59597 returned Unavailable: failed to connect to all addresses
Mar 24 03:01:27 No trace event is collected after 5 attempt(s). Perhaps, you want to try again (with more attempts?).
Mar 24 03:01:27 Tip: increase number of attempts with --num_tracing_attempts.
Mar 24 03:01:27 E
Mar 24 03:01:27 ======================================================================
Mar 24 03:01:27 ERROR: test_trace_and_metrics (__main__.ProfilerTest)
Mar 24 03:01:27 ----------------------------------------------------------------------
Mar 24 03:01:27 Traceback (most recent call last):
Mar 24 03:01:27   File "/var/lib/jenkins/workspace/xla/test/test_profiler.py", line 84, in test_trace_and_metrics
Mar 24 03:01:27     delay_ms=1000)
Mar 24 03:01:27   File "/opt/conda/lib/python3.6/site-packages/torch_xla-1.9-py3.6-linux-x86_64.egg/torch_xla/debug/profiler.py", line 85, in trace
Mar 24 03:01:27     options=options)
Mar 24 03:01:27 RuntimeError: No trace event was collected because there were no responses from clients or the responses did not have trace data.
Mar 24 03:01:27 
Mar 24 03:01:27 ----------------------------------------------------------------------
Mar 24 03:01:27 Ran 1 test in 41.294s

🚧 1 ongoing upstream failure:

These were probably caused by upstream breakages that are not fixed yet.

clang-tidy since Mar 22 (d628e67)
- 🔁 rerun

This comment was automatically generated by Dr. CI (expand for details).

Follow this link to opt-out of these comments for your Pull Requests.

Please report bugs/suggestions to the (internal) Dr. CI Users group.

[ghstack-poisoned]

ghstack-source-id: 1c061f6ea7ab62f80bbb891b4dd98ec0771ea6a5 Pull Request resolved: #52881

[ghstack-poisoned]

ghstack-source-id: 08935ee6a77fb3676773df1726c3b552e462afd9 Pull Request resolved: #52881

[ghstack-poisoned]

ghstack-source-id: 86e0e04ffca5d9a40b0f5d3d43dc05a8610b8f3f Pull Request resolved: #52881

[ghstack-poisoned]

Follow-up work: 1. Add complex support for the unary ops, and other registered ops 2. Extensive TS support for complex lists and dicts 3. Test all emitXYZ for all XYZ in `ir_emitter.cpp` (currently only emitConst, emitValueToTensor are tested). e.g., test loops etc. [ghstack-poisoned]

**This PR adds:** 1. logic to parse complex constants (complex literals of the form `bj`) 2. logic to parse complex lists 3. support for complex constructors: `complex(tensor/int/float/bool, tensor/int/float/bool)` 4. partial complex support for onnx (num -> tensor / tensor -> num) 5. Limited operator support - `add`, `sub`, `mul`, `torch.tensor`, `torch.as_tensor` **Follow-up work:** 1. Add complex support for unary and other registered ops. 2. support complex constructor with string as input (this is supported in Python eager mode). 2. Test all emitXYZ for all XYZ in `ir_emitter.cpp` (currently only emitConst, emitValueToTensor are tested). e.g., test loops etc. [ghstack-poisoned]

SplitInfinity

Nits

torch/csrc/jit/frontend/function_schema_parser.cpp

torch/csrc/jit/ir/ir.cpp

torch/testing/_internal/common_methods_invocations.py

torch/package/_mock.py

torch/csrc/jit/runtime/register_prim_ops_fulljit.cpp

torch/csrc/jit/frontend/ir_emitter.cpp

**This PR adds:** 1. logic to parse complex constants (complex literals of the form `bj`) 2. logic to parse complex lists 3. support for complex constructors: `complex(tensor/int/float/bool, tensor/int/float/bool)` 4. partial complex support for onnx (num -> tensor / tensor -> num) 5. Limited operator support - `add`, `sub`, `mul`, `torch.tensor`, `torch.as_tensor` **Follow-up work:** 1. Add complex support for unary and other registered ops. 2. support complex constructor with string as input (this is supported in Python eager mode). 2. Test all emitXYZ for all XYZ in `ir_emitter.cpp` (currently only emitConst, emitValueToTensor are tested). e.g., test loops etc. [ghstack-poisoned]

**This PR adds:** 1. logic to parse complex constants (complex literals of the form `bj`) 2. logic to parse complex lists 3. support for complex constructors: `complex(tensor/int/float/bool, tensor/int/float/bool)` 4. Limited operator support - `add`, `sub`, `mul`, `torch.tensor`, `torch.as_tensor` **Follow-up work:** 1. Add complex support for unary and other registered ops. 2. support complex constructor with string as input (this is supported in Python eager mode). 3. Test all emitXYZ for all XYZ in `ir_emitter.cpp` (currently only emitConst, emitValueToTensor are tested). e.g., test loops etc. 4. onnx doesn't support complex tensors, so we should error out with a clear and descriptive error message. [ghstack-poisoned]

anjali411 · 2021-03-22T21:57:32Z

@BowenBao @SplitInfinity Removed all the onnx related changes from this PR. I'll add proper error checking for onnx in a follow-up PR!

**This PR adds:** 1. logic to parse complex constants (complex literals of the form `bj`) 2. logic to parse complex lists 3. support for complex constructors: `complex(tensor/int/float/bool, tensor/int/float/bool)` 4. Limited operator support - `add`, `sub`, `mul`, `torch.tensor`, `torch.as_tensor` **Follow-up work:** 1. Add complex support for unary and other registered ops. 2. support complex constructor with string as input (this is supported in Python eager mode). 3. Test all emitXYZ for all XYZ in `ir_emitter.cpp` (currently only emitConst, emitValueToTensor are tested). e.g., test loops etc. 4. onnx doesn't support complex tensors, so we should error out with a clear and descriptive error message. Differential Revision: [D27245059](https://our.internmc.facebook.com/intern/diff/D27245059) [ghstack-poisoned]

SplitInfinity

I can't find any reason to not approve. Please make sure to fix failing tests before landing.

**This PR adds:** 1. logic to parse complex constants (complex literals of the form `bj`) 2. logic to parse complex lists 3. support for complex constructors: `complex(tensor/int/float/bool, tensor/int/float/bool)` 4. Limited operator support - `add`, `sub`, `mul`, `torch.tensor`, `torch.as_tensor` **Follow-up work:** 1. Add complex support for unary and other registered ops. 2. support complex constructor with string as input (this is supported in Python eager mode). 3. Test all emitXYZ for all XYZ in `ir_emitter.cpp` (currently only emitConst, emitValueToTensor are tested). e.g., test loops etc. 4. onnx doesn't support complex tensors, so we should error out with a clear and descriptive error message. Differential Revision: [D27245059](https://our.internmc.facebook.com/intern/diff/D27245059) [ghstack-poisoned]

facebook-github-bot · 2021-03-24T15:13:46Z

@anjali411 merged this pull request in f9ca0d8.

Teach Python TS frontend to parse complex literals

0ea39d1

[ghstack-poisoned]

facebook-github-bot added the cla signed label Feb 25, 2021

facebook-github-bot added the oncall: jit Add this issue/PR to JIT oncall triage queue label Feb 25, 2021

anjali411 changed the title ~~Teach Python TS frontend to parse complex literals~~ [WIP] Teach Python TS frontend to parse complex literals Feb 25, 2021

Update on "[WIP] Teach Python TS frontend to parse complex literals"

8a87e90

[ghstack-poisoned]

anjali411 added a commit that referenced this pull request Feb 25, 2021

Teach Python TS frontend to parse complex literals

5112ed5

ghstack-source-id: 1c061f6ea7ab62f80bbb891b4dd98ec0771ea6a5 Pull Request resolved: #52881

Update on "[WIP] Teach Python TS frontend to parse complex literals"

23d1eb7

[ghstack-poisoned]

anjali411 marked this pull request as draft March 2, 2021 00:16

Update on "[WIP] Teach Python TS frontend to parse complex literals"

5417e49

[ghstack-poisoned]

Update on "[WIP] Teach Python TS frontend to parse complex literals"

4223295

[ghstack-poisoned]

Update on "[WIP] Teach Python TS frontend to parse complex literals"

60f29ca

[ghstack-poisoned]

Update on "[WIP] Teach Python TS frontend to parse complex literals"

0b5feb3

[ghstack-poisoned]

anjali411 added a commit that referenced this pull request Mar 2, 2021

Teach Python TS frontend to parse complex literals

6e03fba

ghstack-source-id: 08935ee6a77fb3676773df1726c3b552e462afd9 Pull Request resolved: #52881

Update on "[WIP] Teach Python TS frontend to parse complex literals"

a2469c5

[ghstack-poisoned]

Update on "[WIP] Teach Python TS frontend to parse complex literals"

bb4600b

[ghstack-poisoned]

anjali411 added a commit that referenced this pull request Mar 2, 2021

Teach Python TS frontend to parse complex literals

7ff4a51

ghstack-source-id: 86e0e04ffca5d9a40b0f5d3d43dc05a8610b8f3f Pull Request resolved: #52881

Update on "[WIP] Teach Python TS frontend to parse complex literals"

aeaf02e

[ghstack-poisoned]

Update on "[WIP] Teach Python TS frontend to parse complex literals"

7e92277

[ghstack-poisoned]

Update on "[WIP] Teach Python TS frontend to parse complex literals"

b04b3f3

[ghstack-poisoned]

Update on "[WIP] Teach Python TS frontend to parse complex literals"

0c61612

[ghstack-poisoned]

anjali411 added 4 commits March 17, 2021 17:21

anjali411 mentioned this pull request Mar 18, 2021

[TS] Support broadcasting a complex value as a list #54285

Closed

SplitInfinity reviewed Mar 19, 2021

View reviewed changes

anjali411 mentioned this pull request Mar 19, 2021

Allow creating SugaredValue for a complex valued IValue and deserialization logic for "infj" and "nanj" global constants #54328

Closed

anjali411 mentioned this pull request Mar 19, 2021

Support torch.complex #53227

Closed

anjali411 added 2 commits March 22, 2021 14:29

anjali411 requested review from BowenBao and SplitInfinity March 22, 2021 21:58

anjali411 added 3 commits March 22, 2021 15:40

anjali411 requested review from SplitInfinity and removed request for SplitInfinity March 23, 2021 15:00

SplitInfinity approved these changes Mar 23, 2021

View reviewed changes

This was referenced Mar 23, 2021

[TS] Add complex support for more ops #54541

Closed

Add support for real and imag tensor attributes #54542

Closed

facebook-github-bot closed this in f9ca0d8 Mar 24, 2021

facebook-github-bot added the Merged label Mar 24, 2021

facebook-github-bot deleted the gh/anjali411/101/head branch March 28, 2021 14:17

anjali411 mentioned this pull request Apr 12, 2021

The jit misinterprets complex values as integers #36368

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Teach Python TS frontend to parse complex literals #52881

Teach Python TS frontend to parse complex literals #52881

anjali411 commented Feb 25, 2021 •

edited

Loading

facebook-github-bot commented Feb 25, 2021 •

edited

Loading

SplitInfinity left a comment

anjali411 commented Mar 22, 2021 •

edited

Loading

SplitInfinity left a comment

facebook-github-bot commented Mar 24, 2021

Teach Python TS frontend to parse complex literals #52881

Teach Python TS frontend to parse complex literals #52881

Conversation

anjali411 commented Feb 25, 2021 • edited Loading

facebook-github-bot commented Feb 25, 2021 • edited Loading

💊 CI failures summary and remediations

🕵️ 1 new failure recognized by patterns

pytorch_xla_linux_bionic_py3_6_clang9_test (1/1)

🚧 1 ongoing upstream failure:

SplitInfinity left a comment

Choose a reason for hiding this comment

anjali411 commented Mar 22, 2021 • edited Loading

SplitInfinity left a comment

Choose a reason for hiding this comment

facebook-github-bot commented Mar 24, 2021

anjali411 commented Feb 25, 2021 •

edited

Loading

facebook-github-bot commented Feb 25, 2021 •

edited

Loading

anjali411 commented Mar 22, 2021 •

edited

Loading