Introduce AT_FORALL_QINTS

[gchanan]

Stack from ghstack:

• Remove unused parameter from FORALL macros and rename STUBS to QINTS. #23340 [WIP] Remove unused parameter from FORALL macros and rename STUBS to QINTS.
• Align AT_FORALL macros with AT_DISPATCH macros. #23339 [WIP] Align AT_FORALL macros with AT_DISPATCH macros.
• Don't redefine unecessary type stub. #23338 [WIP] Don't redefine unecessary type stub.
• Support torch::tensor and at::tensor with bool and BFloat16 dtypes. #23337 [WIP] Support torch::tensor and at::tensor with bool and BFloat16 dtypes.
• Rename AT_FORALL_SCALAR_TYPES_WITH_COMPLEX to AT_FORALL_SCALAR_TYPES_WITH_COMPLEX_AND_STUBS #23336 [WIP] Rename AT_FORALL_SCALAR_TYPES_WITH_COMPLEX to AT_FORALL_SCALAR_TYPES_WITH_COMPLEX_AND_STUBS
• Remove AT_FORALL_SCALAR_TYPES_WITH_COMPLEX_EXCEPT_COMPLEX_HALF, which isn't used anymore. #22932 Remove AT_FORALL_SCALAR_TYPES_WITH_COMPLEX_EXCEPT_COMPLEX_HALF, which isn't used anymore.
• Introduce AT_FORALL_QINTS #22931 Dont allow conversions with QInt.
• Remove cases of AT_FORALL_SCALAR_TYPES_EXCEPT_HALF. #22890 Remove cases of AT_FORALL_SCALAR_TYPES_EXCEPT_HALF.
• C++ at::Tensor, torch::tensor constructors should not accept QInts. #22889 C++ at::Tensor, torch::tensor constructors should not accept QInts.

Differential Revision: D16467985

[jerryzh168]

do we want to cast them to quint8* or uint8*?

[jerryzh168]

I feel uint8* might be more useful for people

[gchanan]

ya, I think that's the correct solution but outside of the scope of this PR, which is just trying to rationalize the AT_FORALL macros. So I'm just trying to match up with this one:

pytorch/aten/src/ATen/Dispatch.h

Lines 256 to 269 in 34f5356

	#define AT_DISPATCH_QINT_TYPES(TYPE, NAME, ...) \
	[&] { \
	const auto& SCALAR_TYPE C10_UNUSED = TYPE; \
	switch (TYPE) { \
	AT_QINT_PRIVATE_CASE_TYPE( \
	at::kQInt8, at::qint8, at::kChar, int8_t, __VA_ARGS__) \
	AT_QINT_PRIVATE_CASE_TYPE( \
	at::kQUInt8, at::quint8, at::kByte, uint8_t, __VA_ARGS__) \
	AT_QINT_PRIVATE_CASE_TYPE( \
	at::kQInt32, at::qint32, at::kInt, int, __VA_ARGS__) \
	default: \
	AT_ERROR(#NAME, " not implemented for '", toString(TYPE), "'"); \
	} \
	}()

so I should have actually called this AT_FORALL_QINT_TYPES, but I fix that up in a later PR in this stack.

[gchanan]

I filed #23440 for addressing that.

[jerryzh168]

LGTM

[facebook-github-bot]

@gchanan merged this pull request in 953459f.