Weinan Liu fa4cb91846 add support for ir scalar literal parsing for inf/-inf/True/False (#163924) ... Currently the ir parser doesn't support parse ir like ``` graph(): %12 : float = prim::Constant[value=-inf]() %13 : float = prim::Constant[value=inf]() %14 : bool = prim::Constant[value=True]() %15 : bool = prim::Constant[value=False]() return (%12) ``` So the python script below will throw error. ``` #!/bin/env python import torch def test(): return [True, False] f = torch.jit.script(test) torch._C._jit_pass_constant_propagation(f.graph) ts_str = f.graph.__repr__() print(ts_str) ts = torch.parse_ir(ts_str) func = torch._C._create_function_from_graph("forward", ts) ret = func() assert ret == [True, False] def test(): return [float("inf"), float("-inf")] f = torch.jit.script(test) torch._C._jit_pass_constant_propagation(f.graph) ts_str = f.graph.__repr__() print(ts_str) ts = torch.parse_ir(ts_str) func = torch._C._create_function_from_graph("forward", ts) ret = func() assert ret == [float("inf"), float("-inf")] ``` I add "inf" and bool cases for IRParser::parseScalarLiteral in irparser.cpp. Pull Request resolved: https://github.com/pytorch/pytorch/pull/163924 Approved by: https://github.com/ezyang		2025-10-27 05:10:21 +00:00
..
alias_analysis.cpp
alias_analysis.h
attributes.cpp
attributes.h	Turn some const strings into constexpr in C++ code (#165203)	2025-10-13 20:25:20 +00:00
constants.cpp
constants.h
graph_node_list.h
graph_utils.cpp
graph_utils.h
ir_views.h
ir.cpp	[ROCm] Fix test_cuda_synchronize failure on ROCm (#164735)	2025-10-07 01:14:24 +00:00
ir.h	[ROCm] Fix test_cuda_synchronize failure on ROCm (#164735)	2025-10-07 01:14:24 +00:00
irparser.cpp	add support for ir scalar literal parsing for inf/-inf/True/False (#163924)	2025-10-27 05:10:21 +00:00
irparser.h
named_value.h
node_hashing.cpp
node_hashing.h
scope.cpp
scope.h	[2/N] Mark unused parameters in C++ code (#165121)	2025-10-15 03:04:39 +00:00
subgraph_matcher.cpp
subgraph_matcher.h
type_hashing.cpp
type_hashing.h