[Fixed Point Arithmetic] Fixed Point Precision Bits and Fixed Point Literals

[leonardchan]

Summary

This diff includes the logic for setting the precision bits for each primary fixed point type in the target info and logic for initializing a fixed point literal.

Fixed point literals are declared using the suffixes

hr: short _Fract
uhr: unsigned short _Fract
r: _Fract
ur: unsigned _Fract
lr: long _Fract
ulr: unsigned long _Fract
hk: short _Accum
uhk: unsigned short _Accum
k: _Accum
uk: unsigned _Accum

Errors are also thrown for illegal literal values

unsigned short _Accum u_short_accum = 256.0uhk;   // expected-error{{the integral part of this literal is too large for this unsigned _Accum type}}

[leonardchan]

formatting

[ebevhan]

You should not define the fixed-point precision as compiler macros at build configure time. The number of fractional bits (the scaling factor) should be added to TargetInfo as target-configurable variables/accessors, and an accessor should be added to ASTContext (we call it 'getFixedPointScale') to fetch the scaling factor of arbitrary fixed-point types.

As I mentioned on the mailing list, we would also like to solve the problem with different scaling factors on the signed and unsigned _Fract types. I'm not sure what the best approach is, but the simplest solution could perhaps be enabled with a target flag.

If the flag is true, the scaling factor of the unsigned _Fract types is the same as the signed ones and the MSB is padding. If the flag is false, the scaling factor of the unsigned _Fract types is one greater than the signed types and all of the bits are used.

---

Overall, the literal parsing code seems very ad-hoc. Fixed-point values are just integers, it should be possible to process them in exact precision instead of relying on host system implementation details (using double) and floating point types.

I could provide patches from our downstream port, but there are a number of hurdles:

• DSP-C does a bunch of extra 'type promotion' when parsing fixed-point literals
• literals in DSP-C cannot contain the 'exponent notation' so the parsing routine cannot handle this
• the routine uses utility functions added to APInt in our LLVM branch

If you are interested I can share anyway so you can see how we have done it.

include/clang/Lex/LiteralSupport.h

72

Is there a reason this is not added as two fields 'isAccum' and 'isFract'?

75

Shouldn't the flags be amended instead?

lib/AST/ASTContext.cpp

1791

See my comments on the _Accum patch.

lib/AST/ASTDumper.cpp

2186

This will not print as a fixed-point number.

lib/AST/Expr.cpp

766

Is this possible given the assertion above?

767

There is no reason to have this switch.

774

'getIntWidth' is likely the wrong accessor to use here. Fixed-point types are technically not ints.

lib/AST/ExprConstant.cpp

9544

Is signed fixed-point overflow actually considered undefined behavior? I'm not familiar with what Embedded-C says about this.

Also, this routine will likely print the wrong number for fixed-point values.

9549

I do not believe ~ is valid on fixed-point types.

lib/AST/StmtPrinter.cpp

1536

This will not print a fixed-point number.

lib/CodeGen/CGExprAgg.cpp

678

This probably goes in the large default case below, as fixed-point types are not aggregates.

lib/CodeGen/CGExprComplex.cpp

451 ↗

(On Diff #147595)

Same as the aggregate case.

lib/CodeGen/CGExprScalar.cpp

1788

This should all use ASTContext accessors.

lib/Sema/SemaExpr.cpp

1294

Fixed-point types.

1295

Braces are not needed.

3336

You cannot parse the literal as a double and then convert it. You cannot control how the value is parsed and you might lose bits.

You should parse it explicitly by hand as a scaled integer large enough to carry the literal and then scale it down and truncate it when you know what the final type will be.

3348

These (bit_width, fbits etc) should be collected from ASTContext after the type has been determined.

3358

It should be possible to do this in a much more concise manner. In our port we do this with a table lookup indexed by the type selection bits.

3434

It does not feel like verifying this is the responsibility of the literal parser.

3441

As I mentioned above, please do not use floating point routines to parse the fixed-point literals.

[leonardchan]

• Moved definition of the fractional/integral bits to TargetInfo. Checks are also performed to make sure any target specific definitions of the F/IBits follow the restrictions set in clause 6.2.6.3 in N1169.
• Added flag -fsame-fbits to force the number of fractional bits in unsigned types to be the same as those of their their corresponding signed types. The default values for all fractional bits is of signed fixed point types are one less than the default values for their unsigned counterparts.
• Implemented custom parsing for fixed point literals and added tests for them. Both decimal and hexadecimal exponentiation are supported.
• I do not have many examples involving assignment to saturated types because this patch does not include conversions yet between fixed point types. This will be in a future patch. The same goes for testing the minimum values for each type because I have not yet implemented folding on binary expressions involving fixed point types.

[leonardchan]

lib/AST/ExprConstant.cpp

9544

I believe clause 4.1.3 states that if FX_FRACT_OVERFLOW or FX_ACCUM_OVERFLOW are set to SAT then overflow behavior on the respective types is saturation. Otherwise, overflow has undefined behavior by default or if the pragma is set to DEFAULT.

lib/Sema/SemaExpr.cpp

3358

Replaced with getters for corresponding type followed by getters for information about the bits.

3434

Right, this is taken care of in the target creation.

[ebevhan]

include/clang/AST/Type.h

6552

This should probably take an APInt or APSInt.

Also, is there a reason to only have it produce unsigned numbers?

include/clang/Basic/LangOptions.def

308

Is it safe to have this as a flag? What about making it a target property?

include/clang/Basic/TargetInfo.h

89

I suspect it's still possible to calculate the ibits based on the fbits, even for _Accum.

Are the unsigned values needed? The fbits for unsigned _Fract are the same as for signed _Fract if SameFBits is set, and +1 otherwise. The same should go for unsigned _Accum, but I don't think it's entirely clear how this affects the integral part.

include/clang/Lex/LiteralSupport.h

116

This should return an APInt. That way you won't run the risk of losing any bits due to overflow or extraneous precision.

lib/AST/ExprConstant.cpp

9542

This doesn't saturate properly. Is that coming in a later patch?

9544

This would probably be simpler if the FixedPointValueToString function could produce signed values.

lib/AST/StmtPrinter.cpp

1540

Format this more like the one above to make fewer lines.

lib/AST/Type.cpp

3990

I think this would be better if it took a SmallString (or SmallVectorImpl) as reference and used that instead of returning a std::string.

Also should probably take an APInt/APSInt.

lib/Lex/LiteralSupport.cpp

1044

This should take an APInt (and use APInts for processing) to store the result in.

It should be possible to calculate exactly how many bits are needed to fit the literal before you start reading the digits. Overflow should not be a problem, but you might get precision loss in the fractional part. The calculation in our version is ceil(log2(10^(noof-digits))) + Scale but ours only handles normal decimal notation (123.456) so it might need to be extended to handle exponents and hex.

lib/Sema/SemaExpr.cpp

1247

I don't see how these semantics work properly. The specification requires that operations be done in the full precision of both types. You cannot convert the types before performing the operation like this, since the operation will not be done in full precision in that case.

The operator semantics of Embedded-C require the operand types of binary operators to be different. It's only when you've performed the operation that you are allowed to convert the result to the resulting type.

1295

I guess you haven't gotten there yet, but this should probably be before handleFloatConversion if you want to handle 'float + _fract' later.

3362

With APInts it should be possible to generalize this code to work with both accum and fract.

[leonardchan]

include/clang/AST/Type.h

6552

Done. It initially took only unsigned numbers since the literals could only be unsigned, but it makes sense to take signed also.

include/clang/Basic/LangOptions.def

308

Can do, but if the point of this option is just to set the Unsigned FBits to their corresponding Signed FBits, then would it be better instead to just ignore this flag altogether and have the target explicitly override the default Unsigned values with the corresponding Signed vals?

include/clang/Basic/TargetInfo.h

89

Similar to the previous comment, if we choose to make SameFBits dependent on/controlled by the target, then I think it would be better for the target to explicitly specify the integral bits since there could be some cases where there may be padding (such as for unsigned _Accum types if this flag is set), and in this case, I think it should be explicit that the bit_width != IBits + FBits.

We also can't fill in that padding for the unsigned _Accum types as an extra integral bit since the standard says that "each signed accum type has at least as many integral bits as its corresponding unsigned accum type".

For the unsigned _Fract values, I think we can get rid of them if we choose to keep the flag instead, but it would no longer apply to unsigned _Accum types since it would allow for extra padding in these types and would conflict with the logic of bit_width == IBits + FBits

For now, I actually think the simplest option is to keep these target properties, but have the target override them individually, with checks to make sure the values adhere to the standard.

lib/AST/ExprConstant.cpp

9542

Handling of saturated types will come in a later patch. For now this is focused on literals for which we cannot specify saturation.

lib/Lex/LiteralSupport.cpp

1044

I see. For the decimal exponent, the number of extra bits needed I think is ceil(Exponent * log2(10)) which requires parsing the exponent before parsing the integral or fractional part, assuming the exponent is positive.

For hex, the number of digits needed just for the integral and fractional part is just 4x the number of digits given in the hex literal + the exponent if it is positive.

With a bit more math we could also trim down the number of digits needed, but I think it's safe just to use the max number of digits needed.

lib/Sema/SemaExpr.cpp

1247

Initially the idea was to convert both sides to fixed point types, then perform standard binary operations between the fixed point types.

For the example, a fract * int would have the int converted to a fixed point type by left shifting it by the scale of the fract, multiplying, then right shifting by the scale again to get the resulting fract. The only unhandled thing is overflow, but the precision of the fract remains the same. The operands would also be casted up beforehand so there was enough space to store the result, which was casted down back to the original fract after performing the right shift by the scale.

Operations between fixed point types would follow a similar process of casting both operands to the higher rank fixed point type, and depending on the operation, more underlying shifting and casting would be done to retain full precision of the higher ranked type.

Though I will admit that I did not realize until now that multiplying a fixed point type by an integer does not require shifting the integer.

1295

A later patch adds a case in handleFloatConversion where we check if one of the operands is an integer and performs a cast from there.

[ebevhan]

include/clang/Basic/TargetInfo.h

89

Is it not the case that bitwidth != IBits + FBits for signed types only? Signed types require a sign bit (which is neither a fractional bit nor an integral bit, according to spec) but unsigned types do not, and if IBits and FBits for the signed and unsigned types are the same, the MSB is simply a padding bit, at least for _Accum (for _Fract everything above the fbits is padding).

My reasoning for keeping the number of configurable values down was to limit the number of twiddly knobs to make the implementation simpler. Granting a lot of freedom is nice, but I suspect that it will be quite hard to get the implementation working properly for every valid configuration. I also don't really see much of a reason for FBits != UFBits in general. I know the spec gives a recommendation to implement it that way, but I think the benefit of normalizing the signed and unsigned representations outweighs the lost bit in the unsigned type.

It's hard to say what the differences are beyond that since I'm not sure how you plan on treating padding bits. In our implementation, padding bits (the MSB in all of the unsigned types for us) after any operation are zeroed.

lib/Basic/TargetInfo.cpp

788

Shouldn't these be ShortAccumIBits >= UnsignedShortAccumIBits etc.?

lib/Lex/LiteralSupport.cpp

736

Is saw_fixed_point_suffix only used for this assertion? Doesn't isFract || isAccum imply saw_fixed_point_suffix?

1048

The function does a lot of overflow checks that I think are superfluous if the required bits are calculated properly. The only overflow check that should be needed is the one at the very bottom that truncates the APInt to the final size.

1064

This whole block is very diligent but I wonder how common overflow in the exponent is.

I'm unsure if LLVM has a helper function with atoll-like functionality, but if it does it's a lot less code to just use that instead. There's also APInt's fromString, but that seems to assert if the integer doesn't fit in the bitwidth. The required bitwidth can be calculated here, though.

1084

These two calculations look very similar to me. The only difference seems to be that the exponent is multiplied by 4 in the decimal case, and not in the hexadecimal case.

1140

I don't think breaking is dependent on this, it's dependent on whether or not the input is a hex digit.

We've already parsed the exponent further up, so we should be able to have an ExponentBegin that we can use as the loop bound.

(It's the same as SuffixBegin if there is no exponent)

1142

Does FoundExponent matter?

1174

This is technically Exponent.

1183

If you want to verify overflow here, you can check that Val.countLeadingZeroes() >= Scale.

1186

I don't think loops are needed here. BaseShift is essentially Exponent so it should be possible to implement as a pow(radix, abs(BaseShift)) and then either a mul or a div based on whether or not it was positive or negative.

1198

I think the final overflow check is the only one you need.

lib/Sema/SemaExpr.cpp

1247

I see how you've reasoned; this is how C normally works. The fract is of higher rank than int and therefore is the 'common type' of the operation. However, even though it is higher rank there is no guarantee that you can perform the operation without overflowing. And overflow matters here; the spec says that it must be done in the full precision (integral + fractional) of both types.

The only unhandled thing is overflow, but the precision of the fract remains the same. The operands would also be casted up beforehand so there was enough space to store the result, which was casted down back to the original fract after performing the right shift by the scale.

The precision remains the same (and while it doesn't have to be the same to perform an operation, it makes the implementation more regular; things like addition and subtraction 'just work'), but you cannot perform a conversion to fract *before* the operation itself, since if you do, there's nothing to 'cast up'. Casting up is needed for things like fract * fract to prevent overflow, but for fract * int you need to cast to a type that can fit both all values of the int and all values of the fract, and *then* you can cast up before doing the multiplication.

Operations between fixed point types would follow a similar process of casting both operands to the higher rank fixed point type, and depending on the operation, more underlying shifting and casting would be done to retain full precision of the higher ranked type.

This might work, but I feel there could be edge cases. The E-C fixed-point ranks are very odd as they don't reflect reality; short _Accum cannot be considered strictly 'above' long _Fract, but the former has a higher rank than the latter. Depending on how the types are specified for a target, implicit casts between fixed-point types might inadvertantly discard bits, even though the spec says that operations must be done in full precision.

1295

one of the operands is an integer

I assume you meant fixed-point here.

[leonardchan]

include/clang/Basic/TargetInfo.h

89

I see. The implementation would be much simpler this way. The initial idea was to treat the padding bits as "don't care" bits where we would mask only the data bits when doing operations like comparisons that look at the whole integer.

lib/Lex/LiteralSupport.cpp

736

saw_fixed_point_suffix is also used to check isFixedPointLiteral() which in turn is also used to check if not isIntegerLiteral() and not isFloatingLiteral() which are called early on in the constructor.

I could substitute where saw_fixed_point_suffix is set true to either setting isAccum or isFract but thought saving those for iterating through the suffixes would make checking for repeated r or k simpler.

1064

Used an APInt instead. It probably won't occur very often, but I would like to cover extreme cases.

1174

For hex types, the exponent base is 2 but every digit past the radix point implies division by 16, hence using a separate variable and multiplying by 4 afterwards.

1186

There's a chance though that the value represented by pow(radix, abs(BaseShift)) may be too large to fit into an integer which is why used the loop.

1198

I kinda want to keep the exponent check to cover extreme cases, but if it seems like too much I don't mind removing it.

lib/Sema/SemaExpr.cpp

1247

I see, so just to confirm, something like a fract * int would not result in any implicit casting between either operand, but any special arithmetic, like intermediate storage types or saturation handling, would be handled by the underlying IR?

So should really no conversions/implicit type casting should be performed here and instead all handling of arithmetic operations should happen somewhere during the codegen stage?

1295

Yup, my bad

[ebevhan]

lib/Lex/LiteralSupport.cpp

1186

True, but I suspect that can be solved by making BaseShift an APInt with the same bitwidth as Val.

lib/Sema/SemaExpr.cpp

1247

I see, so just to confirm, something like a fract * int would not result in any implicit casting between either operand, but any special arithmetic, like intermediate storage types or saturation handling, would be handled by the underlying IR?

Yes, for operations which require precision that cannot be provided by any of the existing types, there must be an 'invisible' implicit conversion to a type which can represent all of the values of either operand. This conversion cannot be represented in the AST as it is today.

The simplest solution is indeed to not have any implicit cast at all in the AST and resolve these conversions when needed (CodeGen and consteval are the locations I can think of), but ultimately it feels a bit dirty... I think that the best solution AST-wise is to define a completely new type class (perhaps FullPrecisionFixedPointType) that represents a fixed-point type with arbitrary width, scale, signedness and saturation. Then you can define ImplicitCasts to an instance of this type that can fit both the int and the fract. I don't know if adding this is acceptable upstream, though.

I think all of these rules must apply to fixed-fixed operations as well; a short accum * long fract must be done as a type that does not exist, similar to fixed-int. It's not clear how saturation should work here either...

I also noticed now that the spec says in regards to comparison operators, When comparing fixed-point values with fixed-point values or integer values, the values are compared directly; the values of the operands are not converted before the comparison is made. I'm not sure what this means.

3362

Does this (and the GetFixedPointValue with it) handle the case where fbits = scale - 1 for a fract?

[ebevhan]

include/clang/Basic/TargetInfo.h

89

That does work, but for binary operations it means you might need multiple maskings per operation. If you mask after every operation instead, you only need one. In our implementation, the only padding bit we have is the MSB in the unsigned types. We simply insert a 'bit clear' after every operation that produces an unsigned type (with some exceptions).

The E-C spec says that padding bits are 'unspecified', but implementation-wise, they have to be defined to something in order to operate on them. So long as you ensure that it is never possible to observe anything but zero in those bits from a program perspective, most operations just work. Naturally, there are ways to subvert this design. If you have a signed _Fract * and cast it to an unsigned _Fract * and the values happen to be negative, the padding bit will be set and you'll get exciting behavior.

Although... I just realized that the topmost half of the _Fracts in your configuration is all padding. That might make things a bit more complicated for things like signed types... Ideally the sign bit would be extended into the padding for those.

[ebevhan]

lib/AST/ExprConstant.cpp

9552

This looks very suspect to me as well... This might not respect padding on types whose sign bit is not the MSB, such as _Fract. The same goes for the overflow check above.

I think it's quite important that we define the semantics of what padding bits should contain. Probably zeroes for unsigned types and a sexted sign bit for signed types.

[leonardchan]

include/clang/Basic/TargetInfo.h

89

Yeah, I initially had 2 general purpose ideas for handling the padding:

• Zero/sign extend them after all operations that modify the value (+, *, <<, casting to another type, ...). This way we know the sign part of a fixed point type, regardless of how much padding there is would, always extend to the padding and sign bit of the underlying scaled integer.
• Masking only the data and sign bits before all operations that read the value and don't modify it (==, !=, <, ...). This way we only compare the non-padding bits whenever performing operations that require reading all bits in the underlying integer.

I wanted to choose the method that addressed operations fixed point types would least likely be used for, that is would these types be generally used more for arithmetic or comparisons. I chose the second method, but don't have any strong evidence to support this. This is also what the later patches implement but can be reworked.

lib/AST/ExprConstant.cpp

9552

Initially I had the _Accum and their corresponding _Fract types the same widths to avoid casting between _Accums to a different sized _Fract (ie short _Fract to short _Accum), but after thinking about it, it might be simpler in the end to just have the _Fract widths have half as much as the _Accum width and have the fractional bits be one less the width (unless SameFBits is set in which case there will still be only 1 padding).

This would also reduce the number of knobs that would be tweaked for different targets and help cover this case by effectively ignoring the padding bits.

[ebevhan]

include/clang/Basic/TargetInfo.h

89

I see what you mean... I also think it's more likely that the types will be used with arithmetic operators than comparisons. However, I don't have any evidence for it but it feels like it could be dangerous to let 'invalid' values hang around; it's safer to ensure that all values are correct after every operation. In our implementation we do the first method for our unsigned types. I wrote a small example (a FIR filter) to see what the overhead of using unsigned instead of signed was, and the x86 asm went from 20 to 23 instructions, which isn't huge. Granted, we do not use unsigned fixed-point types much (if at all) so overhead there isn't really a problem for us.

I think that one of the important benefits of the first method is that if unsigned fixed-point types with padding always have their padding cleared and the fractional bits are the same, then hardware operations for signed fixed-point types can be reused for the unsigned types.

I'm trying to come up with an example where things can go wrong if padding is not corrected after operations, but I can't seem to do it without invoking overflow, which is UB on both signed and unsigned fixed-point if I'm not mistaken. The E-C spec does not seem to mention anything explicit about what happens when you convert a negative signed fixed-point type to an unsigned one, so I think that's UB as well.

lib/AST/ExprConstant.cpp

9552

Yes, that sounds like a good idea. It is also the case for us that the scale of our _Fract types is the width - 1. This makes many operations that care about sign bits trivial, and conversion isn't that bad, it's just a sign/zero extension which should be no problem so long as the type widths are sensible (_Fract to _Accum would be 8->16, 16->32, 32->64 for the respective sizes).

I guess this means dropping the FractFBits configuration values? I suppose the AccumFBits values cannot be dropped as the specification does not have any restrictions between _Accum and _Fract (I think?).

[ebevhan]

lib/Sema/SemaExpr.cpp

1247

In any case, to clarify, I think there are two paths to consider. Either:

• Add a new type class to the type system that encapsulates an arbitrary-precision fixed-point type that can be used for implicit casts when operating on fixed-point and integer types. This is in my opinion the cleaner solution, since it retains invariants on the types of operators and simplifies any logic that deals with operators; or,
• Leave the operands of these operations uncasted. This is in some way simpler, since it doesn't require adding a whole new type, but it complicates other parts of the code. Anything that wants to deal with fixed-point operators will need to know how to do fixed-point conversion as well, which isn't a very good separation of responsibility IMO. It also breaks the C invariant of operands of arithmetic types being in a common type, which might be surprising to people.

[leonardchan]

include/clang/Basic/TargetInfo.h

89

I may go with your recommendation then since this is what you use and tested and I don't have any strong opinions. Also I think conversion from the negative signed to unsigned also falls under UB unless FX_FRACT/ACCUM_OVERFLOW is specified to SAT.

lib/AST/ExprConstant.cpp

9552

They do not have restrictions concerning fractional bits between _Accum and _Fract , but it is labelled as a recommendation to keep them consistent:

- Each unsigned accum type has the same number of fractional bits as its corresponding unsigned fract type.
- Each signed accum type has the same number of fractional bits as either its corresponding signed fract type or its corresponding unsigned fract type.

lib/Sema/SemaExpr.cpp

1247

I'm actually more of a fan for the second case. Aside, aside from the literal parsing in NumericLieralParser, wouldn't the only other place that would actually need to know about fixed point conversion be ScalarExprEmitter under CodeGen/CGExprScalar.cpp?

It seems that it's this class that creates the binary operations and other code gen classes like CodeGenFunction just make underlying calls to ScalarExprEmitter, so the actual conversion logic may just be contained here. Most of the implicit casting handled under UsualArithmeticConversions seems to be handled by VisitCastExpr under ScalarExprEmitter also, so adding another casting type would in the end just result in another case in the switch statement there, which in turn may just result in another call to ScalarExprEmitter.

I can see how it might be weird at first that these types don't fall under usual arithmetic, but the standard does specify that it wouldn't.

1247

Regarding comparison operators, my guess is that it means during comparison operations specifically, the actual underlying values of each operand are compared instead of having the special type conversions take place. That is, 1.0k != 1 but 1.0k == 128 (assuming scale of 7). If this is the case, we could actually save a few operations not having to do a shift on the integer.

I also can't seem to find a test case used by GCC where they explicitly compare a fixed point type against an integer. Normally, they instead assign the FP literal to an integral type, then compare that against another integer.

I'm referring to CONV_ACCUM_INT in
https://github.com/gcc-mirror/gcc/blob/e11be3ea01eaf8acd8cd86d3f9c427621b64e6b4/gcc/testsuite/gcc.dg/fixed-point/convert.h

[leonardchan]

Removed FractFBits property. Fractional bits for _Fract types are now one less the _Fract width unless SameFBits is specified. In that case, the number of FBits in unsigned _Fract is the same as that of signed _Fract, leaving one padding bit that would've been the sign.

[leonardchan]

• Removed the functions for performing casting between fixed point types to a common type when performing a binary operation since operations should be performed on the types as is. Also this patch is meant more for declaring fixed point literals. Binary operations will come in later patches, but it's good to catch these types of things early.
• Removed fixed_point_validation.c since that was the test for the logic described above.

[ebevhan]

lib/Sema/SemaExpr.cpp

1247

I'm actually more of a fan for the second case. Aside, aside from the literal parsing in NumericLieralParser, wouldn't the only other place that would actually need to know about fixed point conversion be ScalarExprEmitter under CodeGen/CGExprScalar.cpp?

ExprConstant (consteval) would also have to know, since the input expressions would be these 'unbalanced' binary operations. I'm not sure why it would affect literal parsing, though?

Regarding VisitCastExpr; in the first case, I'm not talking about adding a new CastKind, I'm talking about adding a whole new type altogether. This type would be just as much a fixed-point type as the builtin ones, just with a configurable width and scale. Then, something like this:

int * fract

where int is 32 bits and fract is 16 bits Q15, would become

(fract)((FullPrecFixedPoint<32+16, 0+15>)int * (FullPrecFixedPoint<32+16, 0+15>)fract)

The cast on the int is a CK_IntegerToFixedPointCast, and the cast on the fract is a CK_FixedPointCast. All values and operations are self-consistent and fully representable in the AST. Converting to and from a FullPrecFixedPoint type is no different from converting to and from, say, fract. They are both fixed-point types with width and scale; one is just built-in and the other is 'artificial'. The multiplication is performed like any other fixed-point operation, just in a higher width and (possibly higher) scale than either of the operands.

The issue I have with the second case is that the AST is somehow left 'unfinished'. There *are* casts there, but they are just not representable in the AST. In order to represent them, you would need to add these arbitrary-precision types.

1247

Regarding comparison operators, my guess is that it means during comparison operations specifically, the actual underlying values of each operand are compared instead of having the special type conversions take place. That is, 1.0k != 1 but 1.0k == 128 (assuming scale of 7). If this is the case, we could actually save a few operations not having to do a shift on the integer.

Right... That seems incredibly dangerous to me; I really hope this isn't what the spec means. 1.0 is by no means the same thing as 128. On top of that, it means that comparisons between fixed-point and integer can vary depending on the scale of the fixed-point type; this feels really shaky to me. Heck, if SameFBits is false, for a scale of 7, 0.5r == 64, but 0.5ur != 64. It might even be the case that 0.5r != 0.5ur. Absolutely bizarre, and incredibly confusing for programmers!

It might have been done this way to make it easy to inspect the raw bits of a fixed-point number, but why not just do a bit-preserving conversion and compare as an integer in that case?

DSP-C simply prohibits 'ambiguous' type conversions such as these to prevent this confusion from happening.

[leonardchan]

• Removed CK_IntegralToFixedPoint to be added to later patch addressing conversion
• Create CXString for FixedPointLiteral

[leonardchan]

lib/Sema/SemaExpr.cpp

1247

I also noticed that the doc provides a bitsfx function for getting this underlying integer value representation from a fixed point type. Still going to leave the comparisons as what I initially interpreted them as by comparing them normally (1.0k == 1).

1247

Understandable. Initially I thought it was a new cast for converting each fixed point type to this intermediate type. I will add this in a new separate patch that addresses conversions involving fixed point types since I feel this one is already large enough and was meant to address just parsing fixed point literals.

[ebevhan]

Just a couple more comments and then I think it looks good.

We can discuss the conversion and comparison issues in later patches.

include/clang/AST/ASTContext.h

1951

These can probably take QualType directly.

include/clang/Basic/TargetInfo.h

99

Could these and their accessors be called 'Scale' like in ASTContext? Only a consistency nit.

lib/Sema/SemaExpr.cpp

1247

Alright. I still suspect that isn't what the standard intends. You save a few instructions, but the comparison doesn't make sense from a value perspective.

We have an externally written test suite which has a bunch of tests for both DSP-C and E-C, and the E-C tests don't seem to behave the way you've described. I've also tried to use the support in avr-gcc to determine how it works and it doesn't seem to do a bitwise comparison there either.

We can discuss this in later patches.

[leonardchan]

lib/Sema/SemaExpr.cpp

1247

Oh, to clarify, I mean that I am performing the shifting necessary when comparing a fixed point to int or another fixed point. We are still comparing the values and not the bits directly.

[ebevhan]

lib/Sema/SemaExpr.cpp

1247

Ah, okay, I misunderstood. I guess it'll be clearer when we get to those patches.

It might be that the spec means that the values should be compared in the full precision of both operands, just like with arithmetic operators. That does require some kind of conversion, though.

[rjmccall]

include/clang/Basic/TargetInfo.h

90

Sorry for the extremely late review, but this really needs to be renamed. Please remember that other compiler maintainers are not experts in the feature that you happen to be currently implementing; they should be able to look at a variable name and understand what it means, and honestly "fbits" does not mean anything. UnsignedFixedPointHasNoPadding, maybe? Or you could shorten it by switching the polarity.

include/clang/Driver/Options.td

899

Okay, please also remember that *users* are not deeply embedded into the lore and wisdom of your implementation group. :) This is not an acceptable user-facing option name; please rename it to something with "fixed-point" in the name.

Also, if this is just for testing (it seems a bit early to decide that users will need an ability to override their target about this?), maybe it should just be a -cc1 option.

[leonardchan]

include/clang/Basic/TargetInfo.h

90

Made a new diff: https://reviews.llvm.org/D48727.

Opted for renaming to PaddingOnUnsignedFixedPoint.

include/clang/Driver/Options.td

899

Renamed to -fpadding-on-unsigned-fixed-point in https://reviews.llvm.org/D48727

[erichkeane]

See this bug here: https://bugs.llvm.org/show_bug.cgi?id=38161

This patch doesn't seem to properly consider integers, and ignores octals, which I suspect shouldn't be the case.

[leonardchan]

In D46915#1162038, @erichkeane wrote:

See this bug here: https://bugs.llvm.org/show_bug.cgi?id=38161

This patch doesn't seem to properly consider integers, and ignores octals, which I suspect shouldn't be the case.

I just requested an account on bugs.llvm.org and will respond here for now until I can comment on the thread there.

So the syntax for fixed point literals should be the same as that for floating point constants according to clause 6.4.4.2a in the embedded C spec (http://www.open-std.org/jtc1/sc22/wg14/www/docs/n1169.pdf). Integers and octal literals should not be supported since floating point literals can only be declared in decimal or hexadecimal notation.

I can see how the error messages don't seem clear and portray this, and should eventually change it so if the user does provide an integer/octal literal an error along the lines of "fixed point literals cannot be declared as integers" should be thrown instead.

The assertion is also something that I will address since this should not be thrown.

To sidetrack a bit also, we are limiting fixed point types to just C for now (https://reviews.llvm.org/D46084) so using auto to complete the type is not supported. This is until we come up with an itanium mangling for fixed point types (https://github.com/itanium-cxx-abi/cxx-abi/issues/56).

[erichkeane]

In D46915#1162095, @leonardchan wrote:

In D46915#1162038, @erichkeane wrote:

See this bug here: https://bugs.llvm.org/show_bug.cgi?id=38161

This patch doesn't seem to properly consider integers, and ignores octals, which I suspect shouldn't be the case.

I just requested an account on bugs.llvm.org and will respond here for now until I can comment on the thread there.

So the syntax for fixed point literals should be the same as that for floating point constants according to clause 6.4.4.2a in the embedded C spec (http://www.open-std.org/jtc1/sc22/wg14/www/docs/n1169.pdf). Integers and octal literals should not be supported since floating point literals can only be declared in decimal or hexadecimal notation.

Then it seems you've missed that part... These literals are allowing integers without issue at the moment.

I can see how the error messages don't seem clear and portray this, and should eventually change it so if the user does provide an integer/octal literal an error along the lines of "fixed point literals cannot be declared as integers" should be thrown instead.

Seems like that could work.

The assertion is also something that I will address since this should not be thrown.

To sidetrack a bit also, we are limiting fixed point types to just C for now (https://reviews.llvm.org/D46084) so using auto to complete the type is not supported. This is until we come up with an itanium mangling for fixed point types (https://github.com/itanium-cxx-abi/cxx-abi/issues/56).

At the moment you are NOT limiting to C. You'll have to add quite a view checks at the moment in order to make that limitation. That said `auto' was for convienience, it had nothign to do with the bug itself.

[leonardchan]

@erichkeane I created a patch at https://reviews.llvm.org/D49327 with the fix for this and added you as a reviewer.

include/clang-c/Index.h

enum CXCursorKind {
   /** OpenMP 4.0 [2.4, Array Section].
    */
   CXCursor_OMPArraySectionExpr           = 147,
 
   /** Represents an @available(...) check.
    */
   CXCursor_ObjCAvailabilityCheckExpr     = 148,
 
-  CXCursor_LastExpr                      = CXCursor_ObjCAvailabilityCheckExpr,
+  /**
+   * Fixed point literal
+   */
+  CXCursor_FixedPointLiteral             = 149,
+
+  CXCursor_LastExpr                      = CXCursor_FixedPointLiteral,
 
   /* Statements */
   CXCursor_FirstStmt                     = 200,
   /**
    * A statement whose specific kind is not exposed via this
    * interface.
    *
    * Unexposed statements have the same operations as any other kind of

include/clang/AST/ASTContext.h

QualType getUnqualifiedObjCPointerType(QualType type) const {
     if (!type.getTypePtr()->isObjCObjectPointerType() ||
         !type.getQualifiers().hasObjCLifetime())
       return type;
     Qualifiers Qs = type.getQualifiers();
     Qs.removeObjCLifetime();
     return getQualifiedType(type.getUnqualifiedType(), Qs);
   }
 
+  unsigned char getFixedPointScale(QualType Ty) const;
+  unsigned char getFixedPointIBits(QualType Ty) const;
+
   DeclarationNameInfo getNameForTemplate(TemplateName Name,
                                          SourceLocation NameLoc) const;
 
   TemplateName getOverloadedTemplateName(UnresolvedSetIterator Begin,
                                          UnresolvedSetIterator End) const;
 
   TemplateName getQualifiedTemplateName(NestedNameSpecifier *NNS,
                                         bool TemplateKeyword,

include/clang/AST/Expr.h

public:
   child_range children() {
     return child_range(child_iterator(), child_iterator());
   }
   const_child_range children() const {
     return const_child_range(const_child_iterator(), const_child_iterator());
   }
 };
 
+class FixedPointLiteral : public Expr, public APIntStorage {
+  SourceLocation Loc;
+  unsigned Scale;
+
+  /// \brief Construct an empty integer literal.
+  explicit FixedPointLiteral(EmptyShell Empty)
+      : Expr(FixedPointLiteralClass, Empty) {}
+
+ public:
+  FixedPointLiteral(const ASTContext &C, const llvm::APInt &V, QualType type,
+                    SourceLocation l, unsigned Scale);
+
+  // Store the int as is without any bit shifting.
+  static FixedPointLiteral *CreateFromRawInt(const ASTContext &C,
+                                             const llvm::APInt &V,
+                                             QualType type, SourceLocation l,
+                                             unsigned Scale);
+
+  SourceLocation getLocStart() const LLVM_READONLY { return Loc; }
+  SourceLocation getLocEnd() const LLVM_READONLY { return Loc; }
+
+  /// \brief Retrieve the location of the literal.
+  SourceLocation getLocation() const { return Loc; }
+
+  void setLocation(SourceLocation Location) { Loc = Location; }
+
+  static bool classof(const Stmt *T) {
+    return T->getStmtClass() == FixedPointLiteralClass;
+  }
+
+  std::string getValueAsString(unsigned Radix) const;
+
+  // Iterators
+  child_range children() {
+    return child_range(child_iterator(), child_iterator());
+  }
+  const_child_range children() const {
+    return const_child_range(const_child_iterator(), const_child_iterator());
+  }
+};
+
 class CharacterLiteral : public Expr {
 public:
   enum CharacterKind {
     Ascii,
     Wide,
     UTF8,
     UTF16,
     UTF32

include/clang/AST/RecursiveASTVisitor.h

DEF_TRAVERSE_STMT(CoyieldExpr, {
   if (!getDerived().shouldVisitImplicitCode()) {
     TRY_TO_TRAVERSE_OR_ENQUEUE_STMT(S->getOperand());
     ShouldVisitChildren = false;
   }
 })
 
 // These literals (all of them) do not need any action.
 DEF_TRAVERSE_STMT(IntegerLiteral, {})
+DEF_TRAVERSE_STMT(FixedPointLiteral, {})
 DEF_TRAVERSE_STMT(CharacterLiteral, {})
 DEF_TRAVERSE_STMT(FloatingLiteral, {})
 DEF_TRAVERSE_STMT(ImaginaryLiteral, {})
 DEF_TRAVERSE_STMT(StringLiteral, {})
 DEF_TRAVERSE_STMT(ObjCStringLiteral, {})
 DEF_TRAVERSE_STMT(ObjCBoxedExpr, {})
 DEF_TRAVERSE_STMT(ObjCArrayLiteral, {})
 DEF_TRAVERSE_STMT(ObjCDictionaryLiteral, {})

include/clang/AST/Type.h

 #include "clang/Basic/ExceptionSpecificationType.h"
 #include "clang/Basic/LLVM.h"
 #include "clang/Basic/Linkage.h"
 #include "clang/Basic/PartialDiagnostic.h"
 #include "clang/Basic/SourceLocation.h"
 #include "clang/Basic/Specifiers.h"
 #include "clang/Basic/Visibility.h"
 #include "llvm/ADT/APInt.h"
+#include "llvm/ADT/APSInt.h"
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/FoldingSet.h"
 #include "llvm/ADT/None.h"
 #include "llvm/ADT/Optional.h"
 #include "llvm/ADT/PointerIntPair.h"
 #include "llvm/ADT/PointerUnion.h"
 #include "llvm/ADT/StringRef.h"
 #include "llvm/ADT/Twine.h"

 }
 
 QualType DecayedType::getPointeeType() const {
   QualType Decayed = getDecayedType();
   (void)AttributedType::stripOuterNullability(Decayed);
   return cast<PointerType>(Decayed)->getPointeeType();
 }
 
+// Get the decimal string representation of a fixed point type, represented
+// as a scaled integer.
+void FixedPointValueToString(SmallVectorImpl<char> &Str,
+                             const llvm::APSInt &Val,
+                             unsigned Scale, unsigned Radix);
+
 } // namespace clang
 
 #endif // LLVM_CLANG_AST_TYPE_H

include/clang/Basic/DiagnosticCommonKinds.td

def ext_old_implicitly_unsigned_long_cxx : ExtWarn<
   "subject to undefined behavior under C++98, interpreting as 'unsigned long'; "
   "this literal will %select{have type 'long long'|be ill-formed}0 "
   "in C++11 onwards">,
   InGroup<CXX11Compat>;
 def ext_clang_enable_if : Extension<"'enable_if' is a clang extension">,
                           InGroup<GccCompat>;
 def ext_clang_diagnose_if : Extension<"'diagnose_if' is a clang extension">,
                             InGroup<GccCompat>;
+def err_too_large_for_fixed_point : Error<
+  "this value is too large for this fixed point type">;
 def err_fixed_point_not_enabled : Error<"compile with "
   "'-ffixed-point' to enable fixed point types">;
 
 // SEH
 def err_seh_expected_handler : Error<
   "expected '__except' or '__finally' block">;
 def err_seh___except_block : Error<
   "%0 only allowed in __except block or filter expression">;

include/clang/Basic/LangOptions.def

 
 ENUM_LANGOPT(ClangABICompat, ClangABI, 4, ClangABI::Latest,
              "version of Clang that we should attempt to be ABI-compatible "
              "with")
 
 COMPATIBLE_VALUE_LANGOPT(FunctionAlignment, 5, 0, "Default alignment for functions")
 
 LANGOPT(FixedPoint, 1, 0, "fixed point types")
+LANGOPT(SameFBits, 1, 0,
+        "unsigned and signed fixed point type having the same number of fractional bits")
 
 #undef LANGOPT
 #undef COMPATIBLE_LANGOPT
 #undef BENIGN_LANGOPT
 #undef ENUM_LANGOPT
 #undef COMPATIBLE_ENUM_LANGOPT
 #undef BENIGN_ENUM_LANGOPT
 #undef VALUE_LANGOPT
 #undef COMPATIBLE_VALUE_LANGOPT
 #undef BENIGN_VALUE_LANGOPT

include/clang/Basic/StmtNodes.td

 def CoroutineBodyStmt : Stmt;
 def CoreturnStmt : Stmt;
 
 // Expressions
 def Expr : Stmt<1>;
 def PredefinedExpr : DStmt<Expr>;
 def DeclRefExpr : DStmt<Expr>;
 def IntegerLiteral : DStmt<Expr>;
+def FixedPointLiteral : DStmt<Expr>;
 def FloatingLiteral : DStmt<Expr>;
 def ImaginaryLiteral : DStmt<Expr>;
 def StringLiteral : DStmt<Expr>;
 def CharacterLiteral : DStmt<Expr>;
 def ParenExpr : DStmt<Expr>;
 def UnaryOperator : DStmt<Expr>;
 def OffsetOfExpr : DStmt<Expr>;
 def UnaryExprOrTypeTraitExpr : DStmt<Expr>;

include/clang/Basic/TargetInfo.h

protected:
   unsigned char IntWidth, IntAlign;
   unsigned char HalfWidth, HalfAlign;
   unsigned char FloatWidth, FloatAlign;
   unsigned char DoubleWidth, DoubleAlign;
   unsigned char LongDoubleWidth, LongDoubleAlign, Float128Align;
   unsigned char LargeArrayMinWidth, LargeArrayAlign;
   unsigned char LongWidth, LongAlign;
   unsigned char LongLongWidth, LongLongAlign;
+
+  // Fixed point bit widths
   unsigned char ShortAccumWidth, ShortAccumAlign;
   unsigned char AccumWidth, AccumAlign;
   unsigned char LongAccumWidth, LongAccumAlign;
   unsigned char ShortFractWidth, ShortFractAlign;
   unsigned char FractWidth, FractAlign;
   unsigned char LongFractWidth, LongFractAlign;
+
+  // If true, unsigned fixed point types have the same number of fractional bits
+  // as their signed counterparts. Otherwise, unsigned fixed point types have
+  // one more fractional bit than its corresponding signed type. This is false
+  // by default.
+  bool SameFBits;
+
+  // Fixed point integral and fractional bit sizes
+  // Saturated types share the same integral/fractional bits as their
+  // corresponding unsaturated types.
+  // For simplicity, the fractional bits in a _Fract type will be one less the
+  // width of that _Fract type. This leaves all signed _Fract types having no
+  // padding and unsigned _Fract types will only have 1 bit of padding after the
+  // sign if SameFBits is set.
+  unsigned char ShortAccumScale;
+  unsigned char AccumScale;
+  unsigned char LongAccumScale;
+
   unsigned char SuitableAlign;
   unsigned char DefaultAlignForAttributeAligned;
   unsigned char MinGlobalAlign;
   unsigned char MaxAtomicPromoteWidth, MaxAtomicInlineWidth;
   unsigned short MaxVectorAlign;
   unsigned short MaxTLSAlign;
   unsigned short SimdDefaultAlign;
   unsigned short NewAlign;
public:
   unsigned getFractWidth() const { return FractWidth; }
   unsigned getFractAlign() const { return FractAlign; }
 
   /// getLongFractWidth/Align - Return the size of 'signed long _Fract' and
   /// 'unsigned long _Fract' for this target, in bits.
   unsigned getLongFractWidth() const { return LongFractWidth; }
   unsigned getLongFractAlign() const { return LongFractAlign; }
 
+  /// getShortAccumScale/IBits - Return the number of fractional/integral bits
+  /// in a 'signed short _Accum' type.
+  unsigned getShortAccumScale() const { return ShortAccumScale; }
+  unsigned getShortAccumIBits() const {
+    return ShortAccumWidth - ShortAccumScale - 1;
+  }
+
+  /// getAccumScale/IBits - Return the number of fractional/integral bits
+  /// in a 'signed _Accum' type.
+  unsigned getAccumScale() const { return AccumScale; }
+  unsigned getAccumIBits() const { return AccumWidth - AccumScale - 1; }
+
+  /// getLongAccumScale/IBits - Return the number of fractional/integral bits
+  /// in a 'signed long _Accum' type.
+  unsigned getLongAccumScale() const { return LongAccumScale; }
+  unsigned getLongAccumIBits() const {
+    return LongAccumWidth - LongAccumScale - 1;
+  }
+
+  /// getUnsignedShortAccumScale/IBits - Return the number of
+  /// fractional/integral bits in a 'unsigned short _Accum' type.
+  unsigned getUnsignedShortAccumScale() const {
+    return SameFBits ? ShortAccumScale : ShortAccumScale + 1;
+  }
+  unsigned getUnsignedShortAccumIBits() const {
+    return SameFBits ? getShortAccumIBits()
+                     : ShortAccumWidth - getUnsignedShortAccumScale();
+  }
+
+  /// getUnsignedAccumScale/IBits - Return the number of fractional/integral
+  /// bits in a 'unsigned _Accum' type.
+  unsigned getUnsignedAccumScale() const {
+    return SameFBits ? AccumScale : AccumScale + 1;
+  }
+  unsigned getUnsignedAccumIBits() const {
+    return SameFBits ? getAccumIBits() : AccumWidth - getUnsignedAccumScale();
+  }
+
+  /// getUnsignedLongAccumScale/IBits - Return the number of fractional/integral
+  /// bits in a 'unsigned long _Accum' type.
+  unsigned getUnsignedLongAccumScale() const {
+    return SameFBits ? LongAccumScale : LongAccumScale + 1;
+  }
+  unsigned getUnsignedLongAccumIBits() const {
+    return SameFBits ? getLongAccumIBits()
+                     : LongAccumWidth - getUnsignedLongAccumScale();
+  }
+
+  /// getShortFractScale - Return the number of fractional bits
+  /// in a 'signed short _Fract' type.
+  unsigned getShortFractScale() const { return ShortFractWidth - 1; }
+
+  /// getFractScale - Return the number of fractional bits
+  /// in a 'signed _Fract' type.
+  unsigned getFractScale() const { return FractWidth - 1; }
+
+  /// getLongFractScale - Return the number of fractional bits
+  /// in a 'signed long _Fract' type.
+  unsigned getLongFractScale() const { return LongFractWidth - 1; }
+
+  /// getUnsignedShortFractScale - Return the number of fractional bits
+  /// in a 'unsigned short _Fract' type.
+  unsigned getUnsignedShortFractScale() const {
+    return SameFBits ? getShortFractScale() : getShortFractScale() + 1;
+  }
+
+  /// getUnsignedFractScale - Return the number of fractional bits
+  /// in a 'unsigned _Fract' type.
+  unsigned getUnsignedFractScale() const {
+    return SameFBits ? getFractScale() : getFractScale() + 1;
+  }
+
+  /// getUnsignedLongFractScale - Return the number of fractional bits
+  /// in a 'unsigned long _Fract' type.
+  unsigned getUnsignedLongFractScale() const {
+    return SameFBits ? getLongFractScale() : getLongFractScale() + 1;
+  }
+
   /// Determine whether the __int128 type is supported on this target.
   virtual bool hasInt128Type() const {
     return (getPointerWidth(0) >= 64) || getTargetOpts().ForceEnableInt128;
   } // FIXME
 
   /// Determine whether _Float16 is supported on this target.
   virtual bool hasLegalHalfType() const { return HasLegalHalfType; }
 
protected:
   virtual enum IntType getPtrDiffTypeV(unsigned AddrSpace) const {
     return PtrDiffType;
   }
   virtual ArrayRef<const char *> getGCCRegNames() const = 0;
   virtual ArrayRef<GCCRegAlias> getGCCRegAliases() const = 0;
   virtual ArrayRef<AddlRegName> getGCCAddlRegNames() const {
     return None;
   }
+
+ private:
+  // Assert the values for the fractional and integral bits for each fixed point
+  // type follow the restrictions given in clause 6.2.6.3 of N1169.
+  void CheckFixedPointBits() const;
 };
 
 }  // end namespace clang
 
 #endif

include/clang/Driver/Options.td

 def fno_force_enable_int128 : Flag<["-"], "fno-force-enable-int128">,
   Group<f_Group>, Flags<[CC1Option]>,
   HelpText<"Disable support for int128_t type">;
 
 def ffixed_point : Flag<["-"], "ffixed-point">, Group<f_Group>,
                    Flags<[CC1Option]>, HelpText<"Enable fixed point types">;
 def fno_fixed_point : Flag<["-"], "fno-fixed-point">, Group<f_Group>,
                       HelpText<"Disable fixed point types">;
+def fsame_fbits : Flag<["-"], "fsame-fbits">, Group<f_Group>,
+                  Flags<[CC1Option]>,
+      HelpText<"Force each unsigned fixed point type to have the same number of fractional bits as its corresponding signed type">;
+def fno_same_fbits : Flag<["-"], "fno-same-fbits">, Group<f_Group>;
 
 // Begin sanitizer flags. These should all be core options exposed in all driver
 // modes.
 let Flags = [CC1Option, CoreOption] in {
 
 def fsanitize_EQ : CommaJoined<["-"], "fsanitize=">, Group<f_clang_Group>,
                    MetaVarName<"<check>">,
                    HelpText<"Turn on runtime checks for various forms of undefined "

include/clang/Lex/LiteralSupport.h

class NumericLiteralParser {
 
   const char *const ThisTokBegin;
   const char *const ThisTokEnd;
   const char *DigitsBegin, *SuffixBegin; // markers
   const char *s; // cursor
 
   unsigned radix;
 
-  bool saw_exponent, saw_period, saw_ud_suffix;
+  bool saw_exponent, saw_period, saw_ud_suffix, saw_fixed_point_suffix;
 
   SmallString<32> UDSuffixBuf;
 
 public:
   NumericLiteralParser(StringRef TokSpelling,
                        SourceLocation TokLoc,
                        Preprocessor &PP);
   bool hadError : 1;
   bool isUnsigned : 1;
   bool isLong : 1;          // This is *not* set for long long.
   bool isLongLong : 1;
   bool isHalf : 1;          // 1.0h
   bool isFloat : 1;         // 1.0f
   bool isImaginary : 1;     // 1.0i
   bool isFloat16 : 1;       // 1.0f16
   bool isFloat128 : 1;      // 1.0q
   uint8_t MicrosoftInteger; // Microsoft suffix extension i8, i16, i32, or i64.
 
+  bool isFract : 1;         // 1.0hr/r/lr/uhr/ur/ulr
+  bool isAccum : 1;         // 1.0hk/k/lk/uhk/uk/ulk
+
+  bool isFixedPointLiteral() const { return saw_fixed_point_suffix; }
+
   bool isIntegerLiteral() const {
-    return !saw_period && !saw_exponent;
+    return !saw_period && !saw_exponent && !isFixedPointLiteral();
   }
   bool isFloatingLiteral() const {
-    return saw_period || saw_exponent;
+    return (saw_period || saw_exponent) && !isFixedPointLiteral();
   }
 
   bool hasUDSuffix() const {
     return saw_ud_suffix;
   }
   StringRef getUDSuffix() const {
     assert(saw_ud_suffix);
     return UDSuffixBuf;
public:
 
   /// GetFloatValue - Convert this numeric literal to a floating value, using
   /// the specified APFloat fltSemantics (specifying float, double, etc).
   /// The optional bool isExact (passed-by-reference) has its value
   /// set to true if the returned APFloat can represent the number in the
   /// literal exactly, and false otherwise.
   llvm::APFloat::opStatus GetFloatValue(llvm::APFloat &Result);
 
+  /// GetFixedPointValue - Convert this numeric literal value into a
+  /// scaled integer that represents this value. Returns true if an overflow
+  /// occurred when calculating the integral part of the scaled integer or
+  /// calculating the digit sequence of the exponent.
+  bool GetFixedPointValue(llvm::APInt &StoreVal, unsigned Scale);
+
 private:
 
   void ParseNumberStartingWithZero(SourceLocation TokLoc);
   void ParseDecimalOrOctalCommon(SourceLocation TokLoc);
 
   static bool isDigitSeparator(char C) { return C == '\''; }
 
   /// Determine whether the sequence of characters [Start, End) contains

lib/AST/ASTContext.cpp

case BuiltinType::Char16:
       Align = Target->getChar16Align();
       break;
     case BuiltinType::Char32:
       Width = Target->getChar32Width();
       Align = Target->getChar32Align();
       break;
     case BuiltinType::UShort:
     case BuiltinType::Short:
       Width = Target->getShortWidth();
       Align = Target->getShortAlign();
       break;
     case BuiltinType::UInt:
     case BuiltinType::Int:
       Width = Target->getIntWidth();
       Align = Target->getIntAlign();
       break;
     case BuiltinType::ULong:

 // Explicitly instantiate this in case a Redeclarable<T> is used from a TU that
 // doesn't include ASTContext.h
 template
 clang::LazyGenerationalUpdatePtr<
     const Decl *, Decl *, &ExternalASTSource::CompleteRedeclChain>::ValueType
 clang::LazyGenerationalUpdatePtr<
     const Decl *, Decl *, &ExternalASTSource::CompleteRedeclChain>::makeValue(
         const clang::ASTContext &Ctx, Decl *Value);
+
+unsigned char ASTContext::getFixedPointScale(QualType Ty) const {
+  assert(Ty->isFixedPointType());
+
+  const auto *BT = Ty->getAs<BuiltinType>();
+  const TargetInfo &Target = getTargetInfo();
+  switch (BT->getKind()) {
+    default:
+      llvm_unreachable("Not a fixed point type!");
+    case BuiltinType::ShortAccum:
+    case BuiltinType::SatShortAccum:
+      return Target.getShortAccumScale();
+    case BuiltinType::Accum:
+    case BuiltinType::SatAccum:
+      return Target.getAccumScale();
+    case BuiltinType::LongAccum:
+    case BuiltinType::SatLongAccum:
+      return Target.getLongAccumScale();
+    case BuiltinType::UShortAccum:
+    case BuiltinType::SatUShortAccum:
+      return Target.getUnsignedShortAccumScale();
+    case BuiltinType::UAccum:
+    case BuiltinType::SatUAccum:
+      return Target.getUnsignedAccumScale();
+    case BuiltinType::ULongAccum:
+    case BuiltinType::SatULongAccum:
+      return Target.getUnsignedLongAccumScale();
+    case BuiltinType::ShortFract:
+    case BuiltinType::SatShortFract:
+      return Target.getShortFractScale();
+    case BuiltinType::Fract:
+    case BuiltinType::SatFract:
+      return Target.getFractScale();
+    case BuiltinType::LongFract:
+    case BuiltinType::SatLongFract:
+      return Target.getLongFractScale();
+    case BuiltinType::UShortFract:
+    case BuiltinType::SatUShortFract:
+      return Target.getUnsignedShortFractScale();
+    case BuiltinType::UFract:
+    case BuiltinType::SatUFract:
+      return Target.getUnsignedFractScale();
+    case BuiltinType::ULongFract:
+    case BuiltinType::SatULongFract:
+      return Target.getUnsignedLongFractScale();
+  }
+}
+
+unsigned char ASTContext::getFixedPointIBits(QualType Ty) const {
+  assert(Ty->isFixedPointType());
+
+  const auto *BT = Ty->getAs<BuiltinType>();
+  const TargetInfo &Target = getTargetInfo();
+  switch (BT->getKind()) {
+    default:
+      llvm_unreachable("Not a fixed point type!");
+    case BuiltinType::ShortAccum:
+    case BuiltinType::SatShortAccum:
+      return Target.getShortAccumIBits();
+    case BuiltinType::Accum:
+    case BuiltinType::SatAccum:
+      return Target.getAccumIBits();
+    case BuiltinType::LongAccum:
+    case BuiltinType::SatLongAccum:
+      return Target.getLongAccumIBits();
+    case BuiltinType::UShortAccum:
+    case BuiltinType::SatUShortAccum:
+      return Target.getUnsignedShortAccumIBits();
+    case BuiltinType::UAccum:
+    case BuiltinType::SatUAccum:
+      return Target.getUnsignedAccumIBits();
+    case BuiltinType::ULongAccum:
+    case BuiltinType::SatULongAccum:
+      return Target.getUnsignedLongAccumIBits();
+    case BuiltinType::ShortFract:
+    case BuiltinType::SatShortFract:
+    case BuiltinType::Fract:
+    case BuiltinType::SatFract:
+    case BuiltinType::LongFract:
+    case BuiltinType::SatLongFract:
+    case BuiltinType::UShortFract:
+    case BuiltinType::SatUShortFract:
+    case BuiltinType::UFract:
+    case BuiltinType::SatUFract:
+    case BuiltinType::ULongFract:
+    case BuiltinType::SatULongFract:
+      return 0;
+  }
+}

lib/AST/ASTDumper.cpp

public:
 
     // Exprs
     void VisitExpr(const Expr *Node);
     void VisitCastExpr(const CastExpr *Node);
     void VisitDeclRefExpr(const DeclRefExpr *Node);
     void VisitPredefinedExpr(const PredefinedExpr *Node);
     void VisitCharacterLiteral(const CharacterLiteral *Node);
     void VisitIntegerLiteral(const IntegerLiteral *Node);
+    void VisitFixedPointLiteral(const FixedPointLiteral *Node);
     void VisitFloatingLiteral(const FloatingLiteral *Node);
     void VisitStringLiteral(const StringLiteral *Str);
     void VisitInitListExpr(const InitListExpr *ILE);
     void VisitArrayInitLoopExpr(const ArrayInitLoopExpr *ILE);
     void VisitArrayInitIndexExpr(const ArrayInitIndexExpr *ILE);
     void VisitUnaryOperator(const UnaryOperator *Node);
     void VisitUnaryExprOrTypeTraitExpr(const UnaryExprOrTypeTraitExpr *Node);
     void VisitMemberExpr(const MemberExpr *Node);

 void ASTDumper::VisitIntegerLiteral(const IntegerLiteral *Node) {
   VisitExpr(Node);
 
   bool isSigned = Node->getType()->isSignedIntegerType();
   ColorScope Color(*this, ValueColor);
   OS << " " << Node->getValue().toString(10, isSigned);
 }
 
+void ASTDumper::VisitFixedPointLiteral(const FixedPointLiteral *Node) {
+  VisitExpr(Node);
+
+  ColorScope Color(*this, ValueColor);
+  OS << " " << Node->getValueAsString(/*Radix=*/10);
+}
+
 void ASTDumper::VisitFloatingLiteral(const FloatingLiteral *Node) {
   VisitExpr(Node);
   ColorScope Color(*this, ValueColor);
   OS << " " << Node->getValueAsApproximateDouble();
 }
 
 void ASTDumper::VisitStringLiteral(const StringLiteral *Str) {
   VisitExpr(Str);

lib/AST/Expr.cpp

IntegerLiteral::Create(const ASTContext &C, const llvm::APInt &V,
   return new (C) IntegerLiteral(C, V, type, l);
 }
 
 IntegerLiteral *
 IntegerLiteral::Create(const ASTContext &C, EmptyShell Empty) {
   return new (C) IntegerLiteral(Empty);
 }
 
+FixedPointLiteral::FixedPointLiteral(const ASTContext &C, const llvm::APInt &V,
+                                     QualType type, SourceLocation l,
+                                     unsigned Scale)
+    : Expr(FixedPointLiteralClass, type, VK_RValue, OK_Ordinary, false, false,
+           false, false),
+      Loc(l), Scale(Scale) {
+  assert(type->isFixedPointType() && "Illegal type in FixedPointLiteral");
+  assert(V.getBitWidth() == C.getTypeInfo(type).Width &&
+         "Fixed point type is not the correct size for constant.");
+  setValue(C, V);
+}
+
+FixedPointLiteral *FixedPointLiteral::CreateFromRawInt(const ASTContext &C,
+                                                       const llvm::APInt &V,
+                                                       QualType type,
+                                                       SourceLocation l,
+                                                       unsigned Scale) {
+  return new (C) FixedPointLiteral(C, V, type, l, Scale);
+}
+
+std::string FixedPointLiteral::getValueAsString(unsigned Radix) const {
+  // Currently the longest decimal number that can be printed is the max for an
+  // unsigned long _Accum: 4294967295.99999999976716935634613037109375
+  // which is 43 characters.
+  SmallString<64> S;
+  FixedPointValueToString(
+      S, llvm::APSInt::getUnsigned(getValue().getZExtValue()), Scale, Radix);
+  return S.str();
+}
+
 FloatingLiteral::FloatingLiteral(const ASTContext &C, const llvm::APFloat &V,
                                  bool isexact, QualType Type, SourceLocation L)
   : Expr(FloatingLiteralClass, Type, VK_RValue, OK_Ordinary, false, false,
          false, false), Loc(L) {
   setSemantics(V.getSemantics());
   FloatingLiteralBits.IsExact = isexact;
   setValue(C, V);
 }
bool Expr::HasSideEffects(const ASTContext &Ctx,
   case TypoExprClass:
   case CXXFoldExprClass:
     llvm_unreachable("shouldn't see dependent / unresolved nodes here");
 
   case DeclRefExprClass:
   case ObjCIvarRefExprClass:
   case PredefinedExprClass:
   case IntegerLiteralClass:
+  case FixedPointLiteralClass:
   case FloatingLiteralClass:
   case ImaginaryLiteralClass:
   case StringLiteralClass:
   case CharacterLiteralClass:
   case OffsetOfExprClass:
   case ImplicitValueInitExprClass:
   case UnaryExprOrTypeTraitExprClass:
   case AddrLabelExprClass:

lib/AST/ExprClassification.cpp

#include "clang/AST/StmtNodes.inc"
   case Expr::CXXNullPtrLiteralExprClass:
   case Expr::ImaginaryLiteralClass:
   case Expr::GNUNullExprClass:
   case Expr::OffsetOfExprClass:
   case Expr::CXXThrowExprClass:
   case Expr::ShuffleVectorExprClass:
   case Expr::ConvertVectorExprClass:
   case Expr::IntegerLiteralClass:
+  case Expr::FixedPointLiteralClass:
   case Expr::CharacterLiteralClass:
   case Expr::AddrLabelExprClass:
   case Expr::CXXDeleteExprClass:
   case Expr::ImplicitValueInitExprClass:
   case Expr::BlockExprClass:
   case Expr::FloatingLiteralClass:
   case Expr::CXXNoexceptExprClass:
   case Expr::CXXScalarValueInitExprClass:

lib/AST/ExprConstant.cpp

public:
   bool VisitUnaryReal(const UnaryOperator *E);
   bool VisitUnaryImag(const UnaryOperator *E);
 
   bool VisitCXXNoexceptExpr(const CXXNoexceptExpr *E);
   bool VisitSizeOfPackExpr(const SizeOfPackExpr *E);
 
   // FIXME: Missing: array subscript of vector, member of vector
 };
+
+class FixedPointExprEvaluator
+    : public ExprEvaluatorBase<FixedPointExprEvaluator> {
+  APValue &Result;
+
+ public:
+  FixedPointExprEvaluator(EvalInfo &info, APValue &result)
+      : ExprEvaluatorBaseTy(info), Result(result) {}
+
+  bool Success(const llvm::APSInt &SI, const Expr *E, APValue &Result) {
+    assert(E->getType()->isFixedPointType() && "Invalid evaluation result.");
+    assert(SI.isSigned() == E->getType()->isSignedFixedPointType() &&
+           "Invalid evaluation result.");
+    assert(SI.getBitWidth() == Info.Ctx.getIntWidth(E->getType()) &&
+           "Invalid evaluation result.");
+    Result = APValue(SI);
+    return true;
+  }
+  bool Success(const llvm::APSInt &SI, const Expr *E) {
+    return Success(SI, E, Result);
+  }
+
+  bool Success(const llvm::APInt &I, const Expr *E, APValue &Result) {
+    assert(E->getType()->isFixedPointType() && "Invalid evaluation result.");
+    assert(I.getBitWidth() == Info.Ctx.getIntWidth(E->getType()) &&
+           "Invalid evaluation result.");
+    Result = APValue(APSInt(I));
+    Result.getInt().setIsUnsigned(E->getType()->isUnsignedFixedPointType());
+    return true;
+  }
+  bool Success(const llvm::APInt &I, const Expr *E) {
+    return Success(I, E, Result);
+  }
+
+  bool Success(uint64_t Value, const Expr *E, APValue &Result) {
+    assert(E->getType()->isFixedPointType() && "Invalid evaluation result.");
+    Result = APValue(Info.Ctx.MakeIntValue(Value, E->getType()));
+    return true;
+  }
+  bool Success(uint64_t Value, const Expr *E) {
+    return Success(Value, E, Result);
+  }
+
+  bool Success(CharUnits Size, const Expr *E) {
+    return Success(Size.getQuantity(), E);
+  }
+
+  bool Success(const APValue &V, const Expr *E) {
+    if (V.isLValue() || V.isAddrLabelDiff()) {
+      Result = V;
+      return true;
+    }
+    return Success(V.getInt(), E);
+  }
+
+  bool ZeroInitialization(const Expr *E) { return Success(0, E); }
+
+  //===--------------------------------------------------------------------===//
+  //                            Visitor Methods
+  //===--------------------------------------------------------------------===//
+
+  bool VisitFixedPointLiteral(const FixedPointLiteral *E) {
+    return Success(E->getValue(), E);
+  }
+
+  bool VisitUnaryOperator(const UnaryOperator *E);
+};
 } // end anonymous namespace
 
 /// EvaluateIntegerOrLValue - Evaluate an rvalue integral-typed expression, and
 /// produce either the integer value or a pointer.
 ///
 /// GCC has a heinous extension which folds casts between pointer types and
 /// pointer-sized integral types. We support this by allowing the evaluation of
 /// an integer rvalue to produce a pointer (represented as an lvalue) instead.

 bool IntExprEvaluator::VisitSizeOfPackExpr(const SizeOfPackExpr *E) {
   return Success(E->getPackLength(), E);
 }
 
 bool IntExprEvaluator::VisitCXXNoexceptExpr(const CXXNoexceptExpr *E) {
   return Success(E->getValue(), E);
 }
 
+bool FixedPointExprEvaluator::VisitUnaryOperator(const UnaryOperator *E) {
+  switch (E->getOpcode()) {
+    default:
+      // Invalid unary operators
+      return Error(E);
+    case UO_Plus:
+      // The result is just the value.
+      return Visit(E->getSubExpr());
+    case UO_Minus: {
+      if (!Visit(E->getSubExpr())) return false;
+      if (!Result.isInt()) return Error(E);
+      const APSInt &Value = Result.getInt();
+      if (Value.isSigned() && Value.isMinSignedValue() && E->canOverflow()) {
+        SmallString<64> S;
+        FixedPointValueToString(S, Value,
+                                Info.Ctx.getTypeInfo(E->getType()).Width,
+                                /*Radix=*/10);
+        Info.CCEDiag(E, diag::note_constexpr_overflow) << S << E->getType();
+        if (Info.noteUndefinedBehavior()) return false;
+      }
+      return Success(-Value, E);
+    }
+    case UO_LNot: {
+      bool bres;
+      if (!EvaluateAsBooleanCondition(E->getSubExpr(), bres, Info))
+        return false;
+      return Success(!bres, E);
+    }
+  }
+}
+
 //===----------------------------------------------------------------------===//
 // Float Evaluation
 //===----------------------------------------------------------------------===//
 
 namespace {
 class FloatExprEvaluator
   : public ExprEvaluatorBase<FloatExprEvaluator> {
   APFloat &Result;
if (E->isGLValue() || T->isFunctionType()) {
     if (!EvaluateFloat(E, F, Info))
       return false;
     Result = APValue(F);
   } else if (T->isAnyComplexType()) {
     ComplexValue C;
     if (!EvaluateComplex(E, C, Info))
       return false;
     C.moveInto(Result);
+  } else if (T->isFixedPointType()) {
+    if (!FixedPointExprEvaluator(Info, Result).Visit(E)) return false;
   } else if (T->isMemberPointerType()) {
     MemberPtr P;
     if (!EvaluateMemberPointer(E, P, Info))
       return false;
     P.moveInto(Result);
     return true;
   } else if (T->isArrayType()) {
     LValue LV;
case Expr::SubstNonTypeTemplateParmExprClass:
     return
       CheckICE(cast<SubstNonTypeTemplateParmExpr>(E)->getReplacement(), Ctx);
 
   case Expr::ParenExprClass:
     return CheckICE(cast<ParenExpr>(E)->getSubExpr(), Ctx);
   case Expr::GenericSelectionExprClass:
     return CheckICE(cast<GenericSelectionExpr>(E)->getResultExpr(), Ctx);
   case Expr::IntegerLiteralClass:
+  case Expr::FixedPointLiteralClass:
   case Expr::CharacterLiteralClass:
   case Expr::ObjCBoolLiteralExprClass:
   case Expr::CXXBoolLiteralExprClass:
   case Expr::CXXScalarValueInitExprClass:
   case Expr::TypeTraitExprClass:
   case Expr::ArrayTypeTraitExprClass:
   case Expr::ExpressionTraitExprClass:
   case Expr::CXXNoexceptExprClass:

lib/AST/ItaniumMangle.cpp

#include "clang/AST/StmtNodes.inc"
   case Expr::TypeTraitExprClass:
   case Expr::ArrayTypeTraitExprClass:
   case Expr::ExpressionTraitExprClass:
   case Expr::VAArgExprClass:
   case Expr::CUDAKernelCallExprClass:
   case Expr::AsTypeExprClass:
   case Expr::PseudoObjectExprClass:
   case Expr::AtomicExprClass:
+  case Expr::FixedPointLiteralClass:
   {
     if (!NullOut) {
       // As bad as this diagnostic is, it's better than crashing.
       DiagnosticsEngine &Diags = Context.getDiags();
       unsigned DiagID = Diags.getCustomDiagID(DiagnosticsEngine::Error,
                                        "cannot yet mangle expression type %0");
       Diags.Report(E->getExprLoc(), DiagID)
         << E->getStmtClassName() << E->getSourceRange();

lib/AST/StmtPrinter.cpp

void StmtPrinter::VisitIntegerLiteral(IntegerLiteral *Node) {
   case BuiltinType::UInt:      OS << 'U'; break;
   case BuiltinType::Long:      OS << 'L'; break;
   case BuiltinType::ULong:     OS << "UL"; break;
   case BuiltinType::LongLong:  OS << "LL"; break;
   case BuiltinType::ULongLong: OS << "ULL"; break;
   }
 }
 
+void StmtPrinter::VisitFixedPointLiteral(FixedPointLiteral *Node) {
+  if (Policy.ConstantsAsWritten && printExprAsWritten(OS, Node, Context))
+    return;
+  OS << Node->getValueAsString(/*Radix=*/10);
+
+  switch (Node->getType()->getAs<BuiltinType>()->getKind()) {
+    default: llvm_unreachable("Unexpected type for fixed point literal!");
+    case BuiltinType::ShortFract:   OS << "hr"; break;
+    case BuiltinType::ShortAccum:   OS << "hk"; break;
+    case BuiltinType::UShortFract:  OS << "uhr"; break;
+    case BuiltinType::UShortAccum:  OS << "uhk"; break;
+    case BuiltinType::Fract:        OS << "r"; break;
+    case BuiltinType::Accum:        OS << "k"; break;
+    case BuiltinType::UFract:       OS << "ur"; break;
+    case BuiltinType::UAccum:       OS << "uk"; break;
+    case BuiltinType::LongFract:    OS << "lr"; break;
+    case BuiltinType::LongAccum:    OS << "lk"; break;
+    case BuiltinType::ULongFract:   OS << "ulr"; break;
+    case BuiltinType::ULongAccum:   OS << "ulk"; break;
+  }
+}
+
 static void PrintFloatingLiteral(raw_ostream &OS, FloatingLiteral *Node,
                                  bool PrintSuffix) {
   SmallString<16> Str;
   Node->getValue().toString(Str);
   OS << Str;
   if (Str.find_first_not_of("-0123456789") == StringRef::npos)
     OS << '.'; // Trailing dot in order to separate from ints.
 

lib/AST/StmtProfile.cpp

 }
 
 void StmtProfiler::VisitIntegerLiteral(const IntegerLiteral *S) {
   VisitExpr(S);
   S->getValue().Profile(ID);
   ID.AddInteger(S->getType()->castAs<BuiltinType>()->getKind());
 }
 
+void StmtProfiler::VisitFixedPointLiteral(const FixedPointLiteral *S) {
+  VisitExpr(S);
+  S->getValue().Profile(ID);
+  ID.AddInteger(S->getType()->castAs<BuiltinType>()->getKind());
+}
+
 void StmtProfiler::VisitCharacterLiteral(const CharacterLiteral *S) {
   VisitExpr(S);
   ID.AddInteger(S->getKind());
   ID.AddInteger(S->getValue());
 }
 
 void StmtProfiler::VisitFloatingLiteral(const FloatingLiteral *S) {
   VisitExpr(S);

lib/AST/Type.cpp

QualType::DestructionKind QualType::isDestructedTypeImpl(QualType type) {
   }
 
   return DK_none;
 }
 
 CXXRecordDecl *MemberPointerType::getMostRecentCXXRecordDecl() const {
   return getClass()->getAsCXXRecordDecl()->getMostRecentNonInjectedDecl();
 }
+
+void clang::FixedPointValueToString(SmallVectorImpl<char> &Str,
+                                    const llvm::APSInt &Val, unsigned Scale,
+                                    unsigned Radix) {
+  llvm::APSInt ScaleVal = llvm::APSInt::getUnsigned(1ULL << Scale);
+  llvm::APSInt IntPart = Val / ScaleVal;
+  llvm::APSInt FractPart = Val % ScaleVal;
+  llvm::APSInt RadixInt = llvm::APSInt::getUnsigned(Radix);
+
+  IntPart.toString(Str, Radix);
+  Str.push_back('.');
+  do {
+    (FractPart * RadixInt / ScaleVal).toString(Str, Radix);
+    FractPart = (FractPart * RadixInt) % ScaleVal;
+  } while (FractPart.getExtValue());
+}

lib/Basic/TargetInfo.cpp

TargetInfo::TargetInfo(const llvm::Triple &T) : TargetOpts(), Triple(T) {
   NoAsmVariants = false;
   HasLegalHalfType = false;
   HasFloat128 = false;
   PointerWidth = PointerAlign = 32;
   BoolWidth = BoolAlign = 8;
   IntWidth = IntAlign = 32;
   LongWidth = LongAlign = 32;
   LongLongWidth = LongLongAlign = 64;
+
+  // Fixed point default bit widths
   ShortAccumWidth = ShortAccumAlign = 16;
   AccumWidth = AccumAlign = 32;
   LongAccumWidth = LongAccumAlign = 64;
-  ShortFractWidth = ShortFractAlign = 16;
-  FractWidth = FractAlign = 32;
-  LongFractWidth = LongFractAlign = 64;
+  ShortFractWidth = ShortFractAlign = 8;
+  FractWidth = FractAlign = 16;
+  LongFractWidth = LongFractAlign = 32;
+
+  // Fixed point default integral and fractional bit sizes
+  // We give the _Accum 1 fewer fractional bits than their corresponding _Fract
+  // types by default to have the same number of fractional bits between _Accum
+  // and _Fract types.
+  SameFBits = false;
+  ShortAccumScale = 7;
+  AccumScale = 15;
+  LongAccumScale = 31;
+
   SuitableAlign = 64;
   DefaultAlignForAttributeAligned = 128;
   MinGlobalAlign = 0;
   // From the glibc documentation, on GNU systems, malloc guarantees 16-byte
   // alignment on 64-bit systems and 8-byte alignment on 32-bit systems. See
   // https://www.gnu.org/software/libc/manual/html_node/Malloc-Examples.html
   if (T.isGNUEnvironment() || T.isWindowsMSVCEnvironment())
     NewAlign = Triple.isArch64Bit() ? 128 : Triple.isArch32Bit() ? 64 : 0;
if (Opts.OpenCL) {
 
     HalfFormat = &llvm::APFloat::IEEEhalf();
     FloatFormat = &llvm::APFloat::IEEEsingle();
     LongDoubleFormat = &llvm::APFloat::IEEEquad();
   }
 
   if (Opts.NewAlignOverride)
     NewAlign = Opts.NewAlignOverride * getCharWidth();
+
+  // Each unsigned fixed point type has the same number of fractional bits as
+  // its corresponding signed type.
+  SameFBits |= Opts.SameFBits;
+  CheckFixedPointBits();
 }
 
 bool TargetInfo::initFeatureMap(
     llvm::StringMap<bool> &Features, DiagnosticsEngine &Diags, StringRef CPU,
     const std::vector<std::string> &FeatureVec) const {
   for (const auto &F : FeatureVec) {
     StringRef Name = F;
     // Apply the feature via the target.
case '*': // Ignore for choosing register preferences.
       break;  // Pass them.
     }
 
     Name++;
   }
 
   return true;
 }
+
+void TargetInfo::CheckFixedPointBits() const {
+  // Check that the number of fractional and integral bits (and maybe sign) can
+  // fit into the bits given for a fixed point type.
+  assert(ShortAccumScale + getShortAccumIBits() + 1 <= ShortAccumWidth);
+  assert(AccumScale + getAccumIBits() + 1 <= AccumWidth);
+  assert(LongAccumScale + getLongAccumIBits() + 1 <= LongAccumWidth);
+  assert(getUnsignedShortAccumScale() + getUnsignedShortAccumIBits() <=
+         ShortAccumWidth);
+  assert(getUnsignedAccumScale() + getUnsignedAccumIBits() <= AccumWidth);
+  assert(getUnsignedLongAccumScale() + getUnsignedLongAccumIBits() <=
+         LongAccumWidth);
+
+  assert(getShortFractScale() + 1 <= ShortFractWidth);
+  assert(getFractScale() + 1 <= FractWidth);
+  assert(getLongFractScale() + 1 <= LongFractWidth);
+  assert(getUnsignedShortFractScale() <= ShortFractWidth);
+  assert(getUnsignedFractScale() <= FractWidth);
+  assert(getUnsignedLongFractScale() <= LongFractWidth);
+
+  // Each unsigned fract type has either the same number of fractional bits
+  // as, or one more fractional bit than, its corresponding signed fract type.
+  assert(getShortFractScale() == getUnsignedShortFractScale() ||
+         getShortFractScale() == getUnsignedShortFractScale() - 1);
+  assert(getFractScale() == getUnsignedFractScale() ||
+         getFractScale() == getUnsignedFractScale() - 1);
+  assert(getLongFractScale() == getUnsignedLongFractScale() ||
+         getLongFractScale() == getUnsignedLongFractScale() - 1);
+
+  // When arranged in order of increasing rank (see 6.3.1.3a), the number of
+  // fractional bits is nondecreasing for each of the following sets of
+  // fixed-point types:
+  // - signed fract types
+  // - unsigned fract types
+  // - signed accum types
+  // - unsigned accum types.
+  assert(getLongFractScale() >= getFractScale() &&
+         getFractScale() >= getShortFractScale());
+  assert(getUnsignedLongFractScale() >= getUnsignedFractScale() &&
+         getUnsignedFractScale() >= getUnsignedShortFractScale());
+  assert(LongAccumScale >= AccumScale && AccumScale >= ShortAccumScale);
+  assert(getUnsignedLongAccumScale() >= getUnsignedAccumScale() &&
+         getUnsignedAccumScale() >= getUnsignedShortAccumScale());
+
+  // When arranged in order of increasing rank (see 6.3.1.3a), the number of
+  // integral bits is nondecreasing for each of the following sets of
+  // fixed-point types:
+  // - signed accum types
+  // - unsigned accum types
+  assert(getLongAccumIBits() >= getAccumIBits() &&
+         getAccumIBits() >= getShortAccumIBits());
+  assert(getUnsignedLongAccumIBits() >= getUnsignedAccumIBits() &&
+         getUnsignedAccumIBits() >= getUnsignedShortAccumIBits());
+
+  // Each signed accum type has at least as many integral bits as its
+  // corresponding unsigned accum type.
+  assert(getShortAccumIBits() >= getUnsignedShortAccumIBits());
+  assert(getAccumIBits() >= getUnsignedAccumIBits());
+  assert(getLongAccumIBits() >= getUnsignedLongAccumIBits());
+}

lib/Basic/Targets.cpp

TargetInfo::CreateTargetInfo(DiagnosticsEngine &Diags,
 
   Target->setSupportedOpenCLOpts();
   Target->setOpenCLExtensionOpts();
   Target->setMaxAtomicWidth();
 
   if (!Target->validateTarget(Diags))
     return nullptr;
 
+  Target->CheckFixedPointBits();
+
   return Target.release();
 }

lib/CodeGen/CGExprAgg.cpp

static Expr *findPeephole(Expr *op, CastKind kind) {
   }
 }
 
 void AggExprEmitter::VisitCastExpr(CastExpr *E) {
   if (const auto *ECE = dyn_cast<ExplicitCastExpr>(E))
     CGF.CGM.EmitExplicitCastExprType(ECE, &CGF);
   switch (E->getCastKind()) {
   case CK_Dynamic: {
     // FIXME: Can this actually happen? We have no test coverage for it.
     assert(isa<CXXDynamicCastExpr>(E) && "CK_Dynamic without a dynamic_cast?");
     LValue LV = CGF.EmitCheckedLValue(E->getSubExpr(),
                                       CodeGenFunction::TCK_Load);
     // FIXME: Do we also need to handle property references here?
     if (LV.isSimple())
       CGF.EmitDynamicCast(LV.getAddress(), cast<CXXDynamicCastExpr>(E));
     else
       CGF.CGM.ErrorUnsupported(E, "non-simple lvalue dynamic_cast");
-    
     if (!Dest.isIgnored())
       CGF.CGM.ErrorUnsupported(E, "lvalue dynamic_cast with a destination");
     break;
   }
-      
   case CK_ToUnion: {
     // Evaluate even if the destination is ignored.
     if (Dest.isIgnored()) {
       CGF.EmitAnyExpr(E->getSubExpr(), AggValueSlot::ignored(),
                       /*ignoreResult=*/true);
       break;
     }
 

lib/CodeGen/CGExprScalar.cpp

public:
   Value *VisitUnaryCoawait(const UnaryOperator *E) {
     return Visit(E->getSubExpr());
   }
 
   // Leaves.
   Value *VisitIntegerLiteral(const IntegerLiteral *E) {
     return Builder.getInt(E->getValue());
   }
+  Value *VisitFixedPointLiteral(const FixedPointLiteral *E) {
+    return Builder.getInt(E->getValue());
+  }
   Value *VisitFloatingLiteral(const FloatingLiteral *E) {
     return llvm::ConstantFP::get(VMContext, E->getValue());
   }
   Value *VisitCharacterLiteral(const CharacterLiteral *E) {
     return llvm::ConstantInt::get(ConvertType(E->getType()), E->getValue());
   }
   Value *VisitObjCBoolLiteralExpr(const ObjCBoolLiteralExpr *E) {
     return llvm::ConstantInt::get(ConvertType(E->getType()), E->getValue());
Value *ScalarExprEmitter::VisitCastExpr(CastExpr *CE) {
   case CK_FloatingToIntegral:
   case CK_FloatingCast:
     return EmitScalarConversion(Visit(E), E->getType(), DestTy,
                                 CE->getExprLoc());
   case CK_BooleanToSignedIntegral:
     return EmitScalarConversion(Visit(E), E->getType(), DestTy,
                                 CE->getExprLoc(),
                                 /*TreatBooleanAsSigned=*/true);
   case CK_IntegralToBoolean:
     return EmitIntToBoolConversion(Visit(E));
   case CK_PointerToBoolean:
     return EmitPointerToBoolConversion(Visit(E), E->getType());
   case CK_FloatingToBoolean:
     return EmitFloatToBoolConversion(Visit(E));
   case CK_MemberPointerToBoolean: {
     llvm::Value *MemPtr = Visit(E);
     const MemberPointerType *MPT = E->getType()->getAs<MemberPointerType>();

lib/Driver/ToolChains/Clang.cpp

#endif
   Args.AddLastArg(CmdArgs, options::OPT_pedantic_errors);
   Args.AddLastArg(CmdArgs, options::OPT_w);
 
   // Fixed point flags
   if (Args.hasFlag(options::OPT_ffixed_point, options::OPT_fno_fixed_point,
                    /*Default=*/false))
     Args.AddLastArg(CmdArgs, options::OPT_ffixed_point);
 
+  if (Args.hasFlag(options::OPT_fsame_fbits,
+                   options::OPT_fno_same_fbits,
+                   /*Default=*/false))
+    Args.AddLastArg(CmdArgs, options::OPT_fsame_fbits);
+
   // Handle -{std, ansi, trigraphs} -- take the last of -{std, ansi}
   // (-ansi is equivalent to -std=c89 or -std=c++98).
   //
   // If a std is supplied, only add -trigraphs if it follows the
   // option.
   bool ImplyVCPPCXXVer = false;
   if (Arg *Std = Args.getLastArg(options::OPT_std_EQ, options::OPT_ansi)) {
     if (Std->getOption().matches(options::OPT_ansi))

lib/Frontend/CompilerInvocation.cpp

#include "clang/Frontend/LangStandards.def"
   Opts.Exceptions = Args.hasArg(OPT_fexceptions);
   Opts.ObjCExceptions = Args.hasArg(OPT_fobjc_exceptions);
   Opts.CXXExceptions = Args.hasArg(OPT_fcxx_exceptions);
 
   // -ffixed-point
   Opts.FixedPoint =
       Args.hasFlag(OPT_ffixed_point, OPT_fno_fixed_point, /*Default=*/false) &&
       !Opts.CPlusPlus;
+  Opts.SameFBits =
+      Args.hasFlag(OPT_fsame_fbits, OPT_fno_same_fbits,
+                   /*Default=*/false) &&
+      Opts.FixedPoint;
 
   // Handle exception personalities
   Arg *A = Args.getLastArg(options::OPT_fsjlj_exceptions,
                            options::OPT_fseh_exceptions,
                            options::OPT_fdwarf_exceptions);
   if (A) {
     const Option &Opt = A->getOption();
     llvm::Triple T(TargetOpts.Triple);

lib/Lex/LiteralSupport.cpp

NumericLiteralParser::NumericLiteralParser(StringRef TokSpelling,
   // the byte at "*end" is both valid and not part of the regex.  Because of
   // this, it doesn't have to check for 'overscan' in various places.
   assert(!isPreprocessingNumberBody(*ThisTokEnd) && "didn't maximally munch?");
 
   s = DigitsBegin = ThisTokBegin;
   saw_exponent = false;
   saw_period = false;
   saw_ud_suffix = false;
+  saw_fixed_point_suffix = false;
   isLong = false;
   isUnsigned = false;
   isLongLong = false;
   isHalf = false;
   isFloat = false;
   isImaginary = false;
   isFloat16 = false;
   isFloat128 = false;
   MicrosoftInteger = 0;
+  isFract = false;
+  isAccum = false;
   hadError = false;
 
   if (*s == '0') { // parse radix
     ParseNumberStartingWithZero(TokLoc);
     if (hadError)
       return;
   } else { // the first digit is non-zero
     radix = 10;
     s = SkipDigits(s);
     if (s == ThisTokEnd) {
       // Done.
     } else {
       ParseDecimalOrOctalCommon(TokLoc);
       if (hadError)
         return;
     }
   }
 
   SuffixBegin = s;
   checkSeparator(TokLoc, s, CSK_AfterDigits);
 
+  // Initial scan to lookahead for fixed point suffix.
+  for (const char *c = s; c != ThisTokEnd; ++c) {
+    if (*c == 'r' || *c == 'k' || *c == 'R' || *c == 'K') {
+      saw_fixed_point_suffix = true;
+      break;
+    }
+  }
+
   // Parse the suffix.  At this point we can classify whether we have an FP or
   // integer constant.
   bool isFPConstant = isFloatingLiteral();
 
   // Loop over all of the characters of the suffix.  If we see something bad,
   // we break out of the loop.
   for (; s != ThisTokEnd; ++s) {
     switch (*s) {
+    case 'R':
+    case 'r':
+      if (isFract || isAccum) break;
+      isFract = true;
+      continue;
+    case 'K':
+    case 'k':
+      if (isFract || isAccum) break;
+      isAccum = true;
+      continue;
     case 'h':      // FP Suffix for "half".
     case 'H':
       // OpenCL Extension v1.2 s9.5 - h or H suffix for half type.
-      if (!PP.getLangOpts().Half) break;
-      if (!isFPConstant) break;  // Error for integer constant.
+      if (!(PP.getLangOpts().Half || PP.getLangOpts().FixedPoint)) break;
+      if (isIntegerLiteral()) break;  // Error for integer constant.
       if (isHalf || isFloat || isLong) break; // HH, FH, LH invalid.
       isHalf = true;
       continue;  // Success.
     case 'f':      // FP Suffix for "float"
     case 'F':
       if (!isFPConstant) break;  // Error for integer constant.
       if (isHalf || isFloat || isLong || isFloat128)
         break; // HF, FF, LF, QF invalid.
if (isValidUDSuffix(PP.getLangOpts(), UDSuffixBuf)) {
         isLong = false;
         isUnsigned = false;
         isLongLong = false;
         isFloat = false;
         isFloat16 = false;
         isHalf = false;
         isImaginary = false;
         MicrosoftInteger = 0;
+        saw_fixed_point_suffix = false;
+        isFract = false;
+        isAccum = false;
       }
 
       saw_ud_suffix = true;
       return;
     }
 
     if (s != ThisTokEnd) {
       // Report an error if there are any.
       PP.Diag(PP.AdvanceToTokenCharacter(TokLoc, SuffixBegin - ThisTokBegin),
               diag::err_invalid_suffix_constant)
           << StringRef(SuffixBegin, ThisTokEnd - SuffixBegin) << isFPConstant;
       hadError = true;
     }
   }
+
+  if (!hadError && saw_fixed_point_suffix) {
+    assert(isFract || isAccum);
+    assert(radix == 16 || radix == 10);
+  }
 }
 
 /// ParseDecimalOrOctalCommon - This method is called for decimal or octal
 /// numbers. It issues an error for illegal digits, and handles floating point
 /// parsing. If it detects a floating point number, the radix is set to 10.
 void NumericLiteralParser::ParseDecimalOrOctalCommon(SourceLocation TokLoc){
   assert((radix == 8 || radix == 10) && "Unexpected radix");
 
if (Str.find('\'') != StringRef::npos) {
     std::remove_copy_if(Str.begin(), Str.end(), std::back_inserter(Buffer),
                         &isDigitSeparator);
     Str = Buffer;
   }
 
   return Result.convertFromString(Str, APFloat::rmNearestTiesToEven);
 }
 
+static inline bool IsExponentPart(char c) {
+  return c == 'p' || c == 'P' || c == 'e' || c == 'E';
+}
+
+bool NumericLiteralParser::GetFixedPointValue(llvm::APInt &StoreVal, unsigned Scale) {
+  assert(radix == 16 || radix == 10);
+
+  // Find how many digits are needed to store the whole literal.
+  unsigned NumDigits = SuffixBegin - DigitsBegin;
+  if (saw_period) --NumDigits;
+
+  // Initial scan of the exponent if it exists
+  bool ExpOverflowOccurred = false;
+  bool NegativeExponent = false;
+  const char *ExponentBegin;
+  uint64_t Exponent = 0;
+  int64_t BaseShift = 0;
+  if (saw_exponent) {
+    const char *Ptr = DigitsBegin;
+
+    while (!IsExponentPart(*Ptr)) ++Ptr;
+    ExponentBegin = Ptr;
+    ++Ptr;
+    NegativeExponent = *Ptr == '-';
+    if (NegativeExponent) ++Ptr;
+
+    unsigned NumExpDigits = SuffixBegin - Ptr;
+    if (alwaysFitsInto64Bits(radix, NumExpDigits)) {
+      llvm::StringRef ExpStr(Ptr, NumExpDigits);
+      llvm::APInt ExpInt(/*numBits=*/64, ExpStr, /*radix=*/10);
+      Exponent = ExpInt.getZExtValue();
+    } else {
+      ExpOverflowOccurred = true;
+    }
+
+    if (NegativeExponent) BaseShift -= Exponent;
+    else BaseShift += Exponent;
+  }
+
+  // Number of bits needed for decimal literal is
+  //   ceil(NumDigits * log2(10))       Integral part
+  // + Scale                            Fractional part
+  // + ceil(Exponent * log2(10))        Exponent
+  // --------------------------------------------------
+  //   ceil((NumDigits + Exponent) * log2(10)) + Scale
+  //
+  // But for simplicity in handling integers, we can round up log2(10) to 4,
+  // making:
+  // 4 * (NumDigits + Exponent) + Scale
+  //
+  // Number of digits needed for hexadecimal literal is
+  //   4 * NumDigits                    Integral part
+  // + Scale                            Fractional part
+  // + Exponent                         Exponent
+  // --------------------------------------------------
+  //   (4 * NumDigits) + Scale + Exponent
+  uint64_t NumBitsNeeded;
+  if (radix == 10)
+    NumBitsNeeded = 4 * (NumDigits + Exponent) + Scale;
+  else
+    NumBitsNeeded = 4 * NumDigits + Exponent + Scale;
+
+  if (NumBitsNeeded > std::numeric_limits<unsigned>::max())
+    ExpOverflowOccurred = true;
+  llvm::APInt Val(static_cast<unsigned>(NumBitsNeeded), 0, /*isSigned=*/false);
+
+  bool FoundDecimal = false;
+
+  int64_t FractBaseShift = 0;
+  const char *End = saw_exponent ? ExponentBegin : SuffixBegin;
+  for (const char *Ptr = DigitsBegin; Ptr < End; ++Ptr) {
+    if (*Ptr == '.') {
+      FoundDecimal = true;
+      continue;
+    }
+
+    // Normal reading of an integer
+    unsigned C = llvm::hexDigitValue(*Ptr);
+    assert(C < radix && "NumericLiteralParser ctor should have rejected this");
+
+    Val *= radix;
+    Val += C;
+
+    if (FoundDecimal)
+      // Keep track of how much we will need to adjust this value by from the
+      // number of digits past the radix point.
+      --FractBaseShift;
+  }
+
+  // For a radix of 16, we will be multiplying by 2 instead of 16.
+  if (radix == 16) FractBaseShift *= 4;
+  BaseShift += FractBaseShift;
+
+  Val <<= Scale;
+
+  uint64_t Base = (radix == 16) ? 2 : 10;
+  if (BaseShift > 0) {
+    for (int64_t i = 0; i < BaseShift; ++i) {
+      Val *= Base;
+    }
+  } else if (BaseShift < 0) {
+    for (int64_t i = BaseShift; i < 0 && !Val.isNullValue(); ++i)
+      Val = Val.udiv(Base);
+  }
+
+  bool IntOverflowOccurred = false;
+  auto MaxVal = llvm::APInt::getMaxValue(StoreVal.getBitWidth());
+  if (Val.getBitWidth() > StoreVal.getBitWidth()) {
+    IntOverflowOccurred |= Val.ugt(MaxVal.zext(Val.getBitWidth()));
+    StoreVal = Val.trunc(StoreVal.getBitWidth());
+  } else if (Val.getBitWidth() < StoreVal.getBitWidth()) {
+    IntOverflowOccurred |= Val.zext(MaxVal.getBitWidth()).ugt(MaxVal);
+    StoreVal = Val.zext(StoreVal.getBitWidth());
+  } else {
+    StoreVal = Val;
+  }
+
+  return IntOverflowOccurred || ExpOverflowOccurred;
+}
+
 /// \verbatim
 ///       user-defined-character-literal: [C++11 lex.ext]
 ///         character-literal ud-suffix
 ///       ud-suffix:
 ///         identifier
 ///       character-literal: [C++11 lex.ccon]
 ///         ' c-char-sequence '
 ///         u' c-char-sequence '
 ///         U' c-char-sequence '
 ///         L' c-char-sequence '
 ///         u8' c-char-sequence ' [C++1z lex.ccon]
 ///       c-char-sequence:
 ///         c-char
 ///         c-char-sequence c-char
 ///       c-char:
 ///         any member of the source character set except the single-quote ',
 ///           backslash \, or new-line character
 ///         escape-sequence
 ///         universal-character-name
 ///       escape-sequence:
 ///         simple-escape-sequence
 ///         octal-escape-sequence
 ///         hexadecimal-escape-sequence
 ///       simple-escape-sequence:
 ///         one of \' \" \? \\ \a \b \f \n \r \t \v
 ///       octal-escape-sequence:
 ///         \ octal-digit
 ///         \ octal-digit octal-digit
 ///         \ octal-digit octal-digit octal-digit
 ///       hexadecimal-escape-sequence:
 ///         \x hexadecimal-digit
 ///         hexadecimal-escape-sequence hexadecimal-digit
 ///       universal-character-name: [C++11 lex.charset]
 ///         \u hex-quad
 ///         \U hex-quad hex-quad
 ///       hex-quad:
 ///         hex-digit hex-digit hex-digit hex-digit
 /// \endverbatim
 ///
 CharLiteralParser::CharLiteralParser(const char *begin, const char *end,
                                      SourceLocation Loc, Preprocessor &PP,
                                      tok::TokenKind kind) {
   // At this point we know that the character matches the regex "(L|u|U)?'.*'".

lib/Sema/SemaExceptionSpec.cpp

CanThrowResult Sema::canThrow(const Expr *E) {
   case Expr::CXXUuidofExprClass:
   case Expr::CharacterLiteralClass:
   case Expr::ExpressionTraitExprClass:
   case Expr::FloatingLiteralClass:
   case Expr::GNUNullExprClass:
   case Expr::ImaginaryLiteralClass:
   case Expr::ImplicitValueInitExprClass:
   case Expr::IntegerLiteralClass:
+  case Expr::FixedPointLiteralClass:
   case Expr::ArrayInitIndexExprClass:
   case Expr::NoInitExprClass:
   case Expr::ObjCEncodeExprClass:
   case Expr::ObjCStringLiteralClass:
   case Expr::ObjCBoolLiteralExprClass:
   case Expr::OpaqueValueExprClass:
   case Expr::PredefinedExprClass:
   case Expr::SizeOfPackExprClass:

lib/Sema/SemaExpr.cpp

QualType Sema::UsualArithmeticConversions(ExprResult &LHS, ExprResult &RHS,
   // For example, "const float" and "float" are equivalent.
   QualType LHSType =
     Context.getCanonicalType(LHS.get()->getType()).getUnqualifiedType();
   QualType RHSType =
     Context.getCanonicalType(RHS.get()->getType()).getUnqualifiedType();
 
   // For conversion purposes, we ignore any atomic qualifier on the LHS.
   if (const AtomicType *AtomicLHS = LHSType->getAs<AtomicType>())
     LHSType = AtomicLHS->getValueType();
 
   // If both types are identical, no conversion is needed.
   if (LHSType == RHSType)
     return LHSType;
 
   // If either side is a non-arithmetic type (e.g. a pointer), we are done.
   // The caller can deal with this (e.g. pointer + int).
   if (!LHSType->isArithmeticType() || !RHSType->isArithmeticType())
if (LHSType->isRealFloatingType() || RHSType->isRealFloatingType())
     return handleFloatConversion(*this, LHS, RHS, LHSType, RHSType,
                                  IsCompAssign);
 
   // Handle GCC complex int extension.
   if (LHSType->isComplexIntegerType() || RHSType->isComplexIntegerType())
     return handleComplexIntConversion(*this, LHS, RHS, LHSType, RHSType,
                                       IsCompAssign);
 
   // Finally, we have two differing integer types.
   return handleIntegerConversion<doIntegralCast, doIntegralCast>
            (*this, LHS, RHS, LHSType, RHSType, IsCompAssign);
 }
 
 
 //===----------------------------------------------------------------------===//
 //  Semantic Analysis for various Expression Types
 //===----------------------------------------------------------------------===//
 
if (Literal.hasUDSuffix()) {
     }
     case LOLR_StringTemplate:
       llvm_unreachable("unexpected literal operator lookup result");
     }
   }
 
   Expr *Res;
 
-  if (Literal.isFloatingLiteral()) {
+  if (Literal.isFixedPointLiteral()) {
+    QualType Ty;
+
+    if (Literal.isAccum) {
+      if (Literal.isHalf) {
+        Ty = Context.ShortAccumTy;
+      } else if (Literal.isLong) {
+        Ty = Context.LongAccumTy;
+      } else {
+        Ty = Context.AccumTy;
+      }
+    } else if (Literal.isFract) {
+      if (Literal.isHalf) {
+        Ty = Context.ShortFractTy;
+      } else if (Literal.isLong) {
+        Ty = Context.LongFractTy;
+      } else {
+        Ty = Context.FractTy;
+      }
+    }
+
+    if (Literal.isUnsigned) Ty = Context.getCorrespondingUnsignedType(Ty);
+
+    bool isSigned = !Literal.isUnsigned;
+    unsigned scale = Context.getFixedPointScale(Ty);
+    unsigned ibits = Context.getFixedPointIBits(Ty);
+    unsigned bit_width = Context.getTypeInfo(Ty).Width;
+
+    llvm::APInt Val(bit_width, 0, isSigned);
+    bool Overflowed = Literal.GetFixedPointValue(Val, scale);
+
+    // Do not use bit_width since some types may have padding like _Fract or
+    // unsigned _Accums if SameFBits is set.
+    auto MaxVal = llvm::APInt::getMaxValue(ibits + scale).zextOrSelf(bit_width);
+    if (Literal.isFract && Val == MaxVal + 1)
+      // Clause 6.4.4 - The value of a constant shall be in the range of
+      // representable values for its type, with exception for constants of a
+      // fract type with a value of exactly 1; such a constant shall denote
+      // the maximal value for the type.
+      --Val;
+    else if (Val.ugt(MaxVal) || Overflowed)
+      Diag(Tok.getLocation(), diag::err_too_large_for_fixed_point);
+
+    Res = FixedPointLiteral::CreateFromRawInt(Context, Val, Ty,
+                                              Tok.getLocation(), scale);
+  } else if (Literal.isFloatingLiteral()) {
     QualType Ty;
     if (Literal.isHalf){
       if (getOpenCLOptions().isEnabled("cl_khr_fp16"))
         Ty = Context.HalfTy;
       else {
         Diag(Tok.getLocation(), diag::err_half_const_requires_fp16);
         return ExprError();
       }
if (Literal.isFixedPointLiteral()) {
     llvm::APInt ResultVal(MaxWidth, 0);
 
     if (Literal.GetIntegerValue(ResultVal)) {
       // If this value didn't fit into uintmax_t, error and force to ull.
       Diag(Tok.getLocation(), diag::err_integer_literal_too_large)
           << /* Unsigned */ 1;
       Ty = Context.UnsignedLongLongTy;
       assert(Context.getTypeSize(Ty) == ResultVal.getBitWidth() &&
              "long long is not intmax_t?");
     } else {
       // If this value fits into a ULL, try to figure out what else it fits into
       // according to the rules of C99 6.4.4.1p5.
 
       // Octal, Hexadecimal, and integers with a U suffix are allowed to
       // be an unsigned int.
       bool AllowUnsigned = Literal.isUnsigned || Literal.getRadix() != 10;
 
       // Check from smallest to largest, picking the smallest type we can.
       unsigned Width = 0;
 
       // Microsoft specific integer suffixes are explicitly sized.
       if (Literal.MicrosoftInteger) {
         if (Literal.MicrosoftInteger == 8 && !Literal.isUnsigned) {
           Width = 8;
if (CondExpr->isTypeDependent() || CondExpr->isValueDependent()) {
     ExprResult CondICE
       = VerifyIntegerConstantExpression(CondExpr, &condEval,
           diag::err_typecheck_choose_expr_requires_constant, false);
     if (CondICE.isInvalid())
       return ExprError();
     CondExpr = CondICE.get();
     CondIsTrue = condEval.getZExtValue();
 
-    // If the condition is > zero, then the AST type is the same as the LSHExpr.
+    // If the condition is > zero, then the AST type is the same as the LHSExpr.
     Expr *ActiveExpr = CondIsTrue ? LHSExpr : RHSExpr;
 
     resType = ActiveExpr->getType();
     ValueDependent = ActiveExpr->isValueDependent();
     VK = ActiveExpr->getValueKind();
     OK = ActiveExpr->getObjectKind();
   }
 

lib/Sema/TreeTransform.h

 }
 
 template<typename Derived>
 ExprResult
 TreeTransform<Derived>::TransformIntegerLiteral(IntegerLiteral *E) {
   return E;
 }
 
-template<typename Derived>
+template <typename Derived>
+ExprResult TreeTransform<Derived>::TransformFixedPointLiteral(
+    FixedPointLiteral *E) {
+  return E;
+}
+
+template<typename Derived>
 ExprResult
 TreeTransform<Derived>::TransformFloatingLiteral(FloatingLiteral *E) {
   return E;
 }
 
 template<typename Derived>
 ExprResult
 TreeTransform<Derived>::TransformImaginaryLiteral(ImaginaryLiteral *E) {

lib/Serialization/ASTReaderStmt.cpp

 }
 
 void ASTStmtReader::VisitIntegerLiteral(IntegerLiteral *E) {
   VisitExpr(E);
   E->setLocation(ReadSourceLocation());
   E->setValue(Record.getContext(), Record.readAPInt());
 }
 
+void ASTStmtReader::VisitFixedPointLiteral(FixedPointLiteral *E) {
+  VisitExpr(E);
+  E->setLocation(ReadSourceLocation());
+  E->setValue(Record.getContext(), Record.readAPInt());
+}
+
 void ASTStmtReader::VisitFloatingLiteral(FloatingLiteral *E) {
   VisitExpr(E);
   E->setRawSemantics(static_cast<Stmt::APFloatSemantics>(Record.readInt()));
   E->setExact(Record.readInt());
   E->setValue(Record.getContext(), Record.readAPFloat(E->getSemantics()));
   E->setLocation(ReadSourceLocation());
 }
 

lib/Serialization/ASTWriterStmt.cpp

void ASTStmtWriter::VisitIntegerLiteral(IntegerLiteral *E) {
 
   if (E->getValue().getBitWidth() == 32) {
     AbbrevToUse = Writer.getIntegerLiteralAbbrev();
   }
 
   Code = serialization::EXPR_INTEGER_LITERAL;
 }
 
+void ASTStmtWriter::VisitFixedPointLiteral(FixedPointLiteral *E) {
+  VisitExpr(E);
+  Record.AddSourceLocation(E->getLocation());
+  Record.AddAPInt(E->getValue());
+  Code = serialization::EXPR_INTEGER_LITERAL;
+}
+
 void ASTStmtWriter::VisitFloatingLiteral(FloatingLiteral *E) {
   VisitExpr(E);
   Record.push_back(E->getRawSemantics());
   Record.push_back(E->isExact());
   Record.AddAPFloat(E->getValue());
   Record.AddSourceLocation(E->getLocation());
   Code = serialization::EXPR_FLOATING_LITERAL;
 }

lib/StaticAnalyzer/Core/ExprEngine.cpp

case Stmt::AsTypeExprClass:
       // Fall through.
 
     // Cases we intentionally don't evaluate, since they don't need
     // to be explicitly evaluated.
     case Stmt::PredefinedExprClass:
     case Stmt::AddrLabelExprClass:
     case Stmt::AttributedStmtClass:
     case Stmt::IntegerLiteralClass:
+    case Stmt::FixedPointLiteralClass:
     case Stmt::CharacterLiteralClass:
     case Stmt::ImplicitValueInitExprClass:
     case Stmt::CXXScalarValueInitExprClass:
     case Stmt::CXXBoolLiteralExprClass:
     case Stmt::ObjCBoolLiteralExprClass:
     case Stmt::ObjCAvailabilityCheckExprClass:
     case Stmt::FloatingLiteralClass:
     case Stmt::NoInitExprClass:

test/Frontend/fixed_point.c

 //CHECK-NEXT: |-VarDecl {{.*}} SatUsF_type 'SatUsF_t':'_Sat unsigned short _Fract'
 //CHECK-NEXT: |-VarDecl {{.*}} SatUF_type 'SatUF_t':'_Sat unsigned _Fract'
 //CHECK-NEXT: |-VarDecl {{.*}} SatUlF_type 'SatUlF_t':'_Sat unsigned long _Fract'
 //CHECK-NEXT: |-VarDecl {{.*}} SatsA_type 'SatsA_t':'_Sat short _Accum'
 //CHECK-NEXT: |-VarDecl {{.*}} SatA_type 'SatA_t':'_Sat _Accum'
 //CHECK-NEXT: |-VarDecl {{.*}} SatlA_type 'SatlA_t':'_Sat long _Accum'
 //CHECK-NEXT: |-VarDecl {{.*}} SatsF_type 'SatsF_t':'_Sat short _Fract'
 //CHECK-NEXT: |-VarDecl {{.*}} SatF_type 'SatF_t':'_Sat _Fract'
-//CHECK-NEXT: `-VarDecl {{.*}} SatlF_type 'SatlF_t':'_Sat long _Fract'
+//CHECK-NEXT: |-VarDecl {{.*}} SatlF_type 'SatlF_t':'_Sat long _Fract'
+
+// Fixed point literal exponent syntax
+_Accum decexp1 = 1.575e1k;
+_Accum decexp2 = 1.575E1k;
+_Accum decexp3 = 1575e-2k;
+_Accum decexp4 = 1575E-2k;
+
+_Accum hexexp1 = 0x0.3p10k;
+_Accum hexexp2 = 0x0.3P10k;
+_Accum hexexp3 = 0x30000p-10k;
+_Accum hexexp4 = 0x30000P-10k;
+
+_Accum zeroexp1 = 1e0k;
+_Accum zeroexp2 = 1e-0k;
+
+//CHECK-NEXT: |-VarDecl {{.*}} decexp1 '_Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Accum' 15.75
+//CHECK-NEXT: |-VarDecl {{.*}} decexp2 '_Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Accum' 15.75
+//CHECK-NEXT: |-VarDecl {{.*}} decexp3 '_Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Accum' 15.75
+//CHECK-NEXT: |-VarDecl {{.*}} decexp4 '_Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Accum' 15.75
+
+//CHECK-NEXT: |-VarDecl {{.*}} hexexp1 '_Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Accum' 192.0
+//CHECK-NEXT: |-VarDecl {{.*}} hexexp2 '_Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Accum' 192.0
+//CHECK-NEXT: |-VarDecl {{.*}} hexexp3 '_Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Accum' 192.0
+//CHECK-NEXT: |-VarDecl {{.*}} hexexp4 '_Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Accum' 192.0
+
+//CHECK-NEXT: |-VarDecl {{.*}} zeroexp1 '_Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Accum' 1.0
+//CHECK-NEXT: |-VarDecl {{.*}} zeroexp2 '_Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Accum' 1.0
+
+// Fixed point literal values
+_Accum literal1 = 2.5k;       // Precise decimal
+_Accum literal2 = 0.0k;       // Zero
+_Accum literal3 = 1.1k;       // Imprecise decimal
+_Accum literal4 = 1.11k;
+_Accum literal5 = 1.111k;
+_Accum literal6 = 1.1111k;
+_Accum literal7 = 1.11111k;   // After some point after the radix, adding any more
+                              // digits to the literal will not result in any
+                              // further precision since the nth digit added may
+                              // be less than the precision that can be
+                              // represented by the fractional bits of the type.
+                              // This results in the same value being stored for
+                              // the type.
+
+//CHECK-NEXT: |-VarDecl {{.*}} literal1 '_Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Accum' 2.5
+//CHECK-NEXT: |-VarDecl {{.*}} literal2 '_Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Accum' 0.0
+//CHECK-NEXT: |-VarDecl {{.*}} literal3 '_Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Accum' 1.0999755859375
+//CHECK-NEXT: |-VarDecl {{.*}} literal4 '_Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Accum' 1.1099853515625
+//CHECK-NEXT: |-VarDecl {{.*}} literal5 '_Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Accum' 1.110992431640625
+//CHECK-NEXT: |-VarDecl {{.*}} literal6 '_Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Accum' 1.111083984375
+//CHECK-NEXT: |-VarDecl {{.*}} literal7 '_Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Accum' 1.111083984375
+
+long _Accum longaccumliteral     = 0.99999999lk;
+long _Accum longaccumliteral2    = 0.999999999lk;
+long _Accum verylongaccumliteral = 0.99999999999999999999999999lk;
+long _Fract longfractliteral     = 0.99999999lr;
+long _Fract longfractliteral2    = 0.999999999lr;
+long _Fract verylongfractliteral = 0.99999999999999999999999999lr;
+
+//CHECK-NEXT: |-VarDecl {{.*}} longaccumliteral 'long _Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} 'long _Accum' 0.999999989755451679229736328125
+//CHECK-NEXT: |-VarDecl {{.*}} longaccumliteral2 'long _Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} 'long _Accum' 0.9999999986030161380767822265625
+//CHECK-NEXT: |-VarDecl {{.*}} verylongaccumliteral 'long _Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} 'long _Accum' 0.9999999995343387126922607421875
+//CHECK-NEXT: |-VarDecl {{.*}} longfractliteral 'long _Fract' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} 'long _Fract' 0.999999989755451679229736328125
+//CHECK-NEXT: |-VarDecl {{.*}} longfractliteral2 'long _Fract' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} 'long _Fract' 0.9999999986030161380767822265625
+//CHECK-NEXT: |-VarDecl {{.*}} verylongfractliteral 'long _Fract' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} 'long _Fract' 0.9999999995343387126922607421875
+
+unsigned _Accum uliteral1 = 2.5uk;    // Unsigned
+_Accum literal8 = -2.5k;              // Negative
+
+//CHECK-NEXT: |-VarDecl {{.*}} uliteral1 'unsigned _Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} 'unsigned _Accum' 2.5
+//CHECK-NEXT: |-VarDecl {{.*}} literal8 '_Accum' cinit
+//CHECK-NEXT:   `-UnaryOperator {{.*}} '_Accum' prefix '-'
+//CHECK-NEXT:     `-FixedPointLiteral {{.*}} '_Accum' 2.5
+
+short _Accum  literalexact1 = 0.9921875hk;  // Exact value
+_Accum        literalexact2 = 0.999969482421875k;
+
+//CHECK-NEXT: |-VarDecl {{.*}} literalexact1 'short _Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} 'short _Accum' 0.9921875
+//CHECK-NEXT: |-VarDecl {{.*}} literalexact2 '_Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Accum' 0.999969482421875
+
+// Unfortunately we do not have enough space to store the exact decimal value of
+// 0.9999999995343387126922607421875 ((1 << 31) - 1), but we can still use a
+// large number of 9s to get the max fractional value.
+long _Accum   long_accum_max = 0.999999999999999999999999999lk;
+
+//CHECK-NEXT: |-VarDecl {{.*}} long_accum_max 'long _Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} 'long _Accum' 0.9999999995343387126922607421875
+
+// Epsilon
+short _Accum  short_accum_eps   = 0.0078125hk;
+short _Accum  short_accum_eps2  = 0.0078124hk;  // Less than epsilon floors to zero
+_Accum        accum_eps         = 0.000030517578125k;
+_Accum        accum_eps2        = 0.000030517578124k;
+long _Accum   long_accum_eps    = 0x1p-31lk;
+long _Accum   long_accum_eps2   = 0x0.99999999p-31lk;
+
+//CHECK-NEXT: |-VarDecl {{.*}} short_accum_eps 'short _Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} 'short _Accum' 0.0078125
+//CHECK-NEXT: |-VarDecl {{.*}} short_accum_eps2 'short _Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} 'short _Accum' 0.0
+//CHECK-NEXT: |-VarDecl {{.*}} accum_eps '_Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Accum' 0.000030517578125
+//CHECK-NEXT: |-VarDecl {{.*}} accum_eps2 '_Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Accum' 0.0
+//CHECK-NEXT: |-VarDecl {{.*}} long_accum_eps 'long _Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} 'long _Accum' 0.0000000004656612873077392578125
+//CHECK-NEXT: |-VarDecl {{.*}} long_accum_eps2 'long _Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} 'long _Accum' 0.0
+
+// Fract literals can be one but evaluate to the respective Fract max
+short _Fract           short_fract_one   = 1.0hr;
+_Fract                 fract_one         = 1.0r;
+long _Fract            long_fract_one    = 1.0lr;
+unsigned short _Fract  u_short_fract_one = 1.0uhr;
+unsigned _Fract        u_fract_one       = 1.0ur;
+unsigned long _Fract   u_long_fract_one  = 1.0ulr;
+
+//CHECK-NEXT: |-VarDecl {{.*}} short_fract_one 'short _Fract' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} 'short _Fract' 0.9921875
+//CHECK-NEXT: |-VarDecl {{.*}} fract_one '_Fract' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Fract' 0.999969482421875
+//CHECK-NEXT: |-VarDecl {{.*}} long_fract_one 'long _Fract' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} 'long _Fract' 0.9999999995343387126922607421875
+
+//CHECK-NEXT: |-VarDecl {{.*}} u_short_fract_one 'unsigned short _Fract' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} 'unsigned short _Fract' 0.99609375
+//CHECK-NEXT: |-VarDecl {{.*}} u_fract_one 'unsigned _Fract' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} 'unsigned _Fract' 0.9999847412109375
+//CHECK-NEXT: |-VarDecl {{.*}} u_long_fract_one 'unsigned long _Fract' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} 'unsigned long _Fract' 0.99999999976716935634613037109375
+
+_Accum literallast = 1.0k;    // One
+
+//CHECK-NEXT: `-VarDecl {{.*}} literallast '_Accum' cinit
+//CHECK-NEXT:   `-FixedPointLiteral {{.*}} '_Accum' 1.0

test/Frontend/fixed_point_bit_widths.c

 
 int size_SsF = sizeof(signed short _Fract);
 int size_SF  = sizeof(signed _Fract);
 int size_SlF = sizeof(signed long _Fract);
 int align_SsF = __alignof(signed short _Fract);
 int align_SF  = __alignof(signed _Fract);
 int align_SlF = __alignof(signed long _Fract);
 
-// CHECK-NEXT: @size_SsF  = {{.*}}global i{{[0-9]+}} 2
-// CHECK-NEXT: @size_SF   = {{.*}}global i{{[0-9]+}} 4
-// CHECK-NEXT: @size_SlF  = {{.*}}global i{{[0-9]+}} 8
-// CHECK-NEXT: @align_SsF = {{.*}}global i{{[0-9]+}} 2
-// CHECK-NEXT: @align_SF  = {{.*}}global i{{[0-9]+}} 4
-// CHECK-NEXT: @align_SlF = {{.*}}global i{{[0-9]+}} 8
+// CHECK-NEXT: @size_SsF  = {{.*}}global i{{[0-9]+}} 1
+// CHECK-NEXT: @size_SF   = {{.*}}global i{{[0-9]+}} 2
+// CHECK-NEXT: @size_SlF  = {{.*}}global i{{[0-9]+}} 4
+// CHECK-NEXT: @align_SsF = {{.*}}global i{{[0-9]+}} 1
+// CHECK-NEXT: @align_SF  = {{.*}}global i{{[0-9]+}} 2
+// CHECK-NEXT: @align_SlF = {{.*}}global i{{[0-9]+}} 4
 
 /* Primary unsigned _Fract */
 
 int size_UsF = sizeof(unsigned short _Fract);
 int size_UF  = sizeof(unsigned _Fract);
 int size_UlF = sizeof(unsigned long _Fract);
 int align_UsF = __alignof(unsigned short _Fract);
 int align_UF  = __alignof(unsigned _Fract);
 int align_UlF = __alignof(unsigned long _Fract);
 
-// CHECK-NEXT: @size_UsF  = {{.*}}global i{{[0-9]+}} 2
-// CHECK-NEXT: @size_UF   = {{.*}}global i{{[0-9]+}} 4
-// CHECK-NEXT: @size_UlF  = {{.*}}global i{{[0-9]+}} 8
-// CHECK-NEXT: @align_UsF = {{.*}}global i{{[0-9]+}} 2
-// CHECK-NEXT: @align_UF  = {{.*}}global i{{[0-9]+}} 4
-// CHECK-NEXT: @align_UlF = {{.*}}global i{{[0-9]+}} 8
+// CHECK-NEXT: @size_UsF  = {{.*}}global i{{[0-9]+}} 1
+// CHECK-NEXT: @size_UF   = {{.*}}global i{{[0-9]+}} 2
+// CHECK-NEXT: @size_UlF  = {{.*}}global i{{[0-9]+}} 4
+// CHECK-NEXT: @align_UsF = {{.*}}global i{{[0-9]+}} 1
+// CHECK-NEXT: @align_UF  = {{.*}}global i{{[0-9]+}} 2
+// CHECK-NEXT: @align_UlF = {{.*}}global i{{[0-9]+}} 4
 
 /* Aliased _Accum */
 
 int size_sA = sizeof(short _Accum);
 int size_A  = sizeof(_Accum);
 int size_lA = sizeof(long _Accum);
 int align_sA = __alignof(short _Accum);
 int align_A  = __alignof(_Accum);

 
 int size_sF = sizeof(short _Fract);
 int size_F  = sizeof(_Fract);
 int size_lF = sizeof(long _Fract);
 int align_sF = __alignof(short _Fract);
 int align_F  = __alignof(_Fract);
 int align_lF = __alignof(long _Fract);
 
-// CHECK-NEXT: @size_sF   = {{.*}}global i{{[0-9]+}} 2
-// CHECK-NEXT: @size_F    = {{.*}}global i{{[0-9]+}} 4
-// CHECK-NEXT: @size_lF   = {{.*}}global i{{[0-9]+}} 8
-// CHECK-NEXT: @align_sF  = {{.*}}global i{{[0-9]+}} 2
-// CHECK-NEXT: @align_F   = {{.*}}global i{{[0-9]+}} 4
-// CHECK-NEXT: @align_lF  = {{.*}}global i{{[0-9]+}} 8
+// CHECK-NEXT: @size_sF   = {{.*}}global i{{[0-9]+}} 1
+// CHECK-NEXT: @size_F    = {{.*}}global i{{[0-9]+}} 2
+// CHECK-NEXT: @size_lF   = {{.*}}global i{{[0-9]+}} 4
+// CHECK-NEXT: @align_sF  = {{.*}}global i{{[0-9]+}} 1
+// CHECK-NEXT: @align_F   = {{.*}}global i{{[0-9]+}} 2
+// CHECK-NEXT: @align_lF  = {{.*}}global i{{[0-9]+}} 4
 
 /* Saturated signed _Accum */
 
 int size_SatSsA = sizeof(_Sat signed short _Accum);
 int size_SatSA  = sizeof(_Sat signed _Accum);
 int size_SatSlA = sizeof(_Sat signed long _Accum);
 int align_SatSsA = __alignof(_Sat signed short _Accum);
 int align_SatSA  = __alignof(_Sat signed _Accum);

 
 int size_SatSsF = sizeof(_Sat signed short _Fract);
 int size_SatSF  = sizeof(_Sat signed _Fract);
 int size_SatSlF = sizeof(_Sat signed long _Fract);
 int align_SatSsF = __alignof(_Sat signed short _Fract);
 int align_SatSF  = __alignof(_Sat signed _Fract);
 int align_SatSlF = __alignof(_Sat signed long _Fract);
 
-// CHECK-NEXT: @size_SatSsF  = {{.*}}global i{{[0-9]+}} 2
-// CHECK-NEXT: @size_SatSF   = {{.*}}global i{{[0-9]+}} 4
-// CHECK-NEXT: @size_SatSlF  = {{.*}}global i{{[0-9]+}} 8
-// CHECK-NEXT: @align_SatSsF = {{.*}}global i{{[0-9]+}} 2
-// CHECK-NEXT: @align_SatSF  = {{.*}}global i{{[0-9]+}} 4
-// CHECK-NEXT: @align_SatSlF = {{.*}}global i{{[0-9]+}} 8
+// CHECK-NEXT: @size_SatSsF  = {{.*}}global i{{[0-9]+}} 1
+// CHECK-NEXT: @size_SatSF   = {{.*}}global i{{[0-9]+}} 2
+// CHECK-NEXT: @size_SatSlF  = {{.*}}global i{{[0-9]+}} 4
+// CHECK-NEXT: @align_SatSsF = {{.*}}global i{{[0-9]+}} 1
+// CHECK-NEXT: @align_SatSF  = {{.*}}global i{{[0-9]+}} 2
+// CHECK-NEXT: @align_SatSlF = {{.*}}global i{{[0-9]+}} 4
 
 /* Saturated unsigned _Fract */
 
 int size_SatUsF = sizeof(_Sat unsigned short _Fract);
 int size_SatUF  = sizeof(_Sat unsigned _Fract);
 int size_SatUlF = sizeof(_Sat unsigned long _Fract);
 int align_SatUsF = __alignof(_Sat unsigned short _Fract);
 int align_SatUF  = __alignof(_Sat unsigned _Fract);
 int align_SatUlF = __alignof(_Sat unsigned long _Fract);
 
-// CHECK-NEXT: @size_SatUsF  = {{.*}}global i{{[0-9]+}} 2
-// CHECK-NEXT: @size_SatUF   = {{.*}}global i{{[0-9]+}} 4
-// CHECK-NEXT: @size_SatUlF  = {{.*}}global i{{[0-9]+}} 8
-// CHECK-NEXT: @align_SatUsF = {{.*}}global i{{[0-9]+}} 2
-// CHECK-NEXT: @align_SatUF  = {{.*}}global i{{[0-9]+}} 4
-// CHECK-NEXT: @align_SatUlF = {{.*}}global i{{[0-9]+}} 8
+// CHECK-NEXT: @size_SatUsF  = {{.*}}global i{{[0-9]+}} 1
+// CHECK-NEXT: @size_SatUF   = {{.*}}global i{{[0-9]+}} 2
+// CHECK-NEXT: @size_SatUlF  = {{.*}}global i{{[0-9]+}} 4
+// CHECK-NEXT: @align_SatUsF = {{.*}}global i{{[0-9]+}} 1
+// CHECK-NEXT: @align_SatUF  = {{.*}}global i{{[0-9]+}} 2
+// CHECK-NEXT: @align_SatUlF = {{.*}}global i{{[0-9]+}} 4
 
 /* Aliased saturated signed _Accum */
 
 int size_SatsA = sizeof(_Sat short _Accum);
 int size_SatA  = sizeof(_Sat _Accum);
 int size_SatlA = sizeof(_Sat long _Accum);
 int align_SatsA = __alignof(_Sat short _Accum);
 int align_SatA  = __alignof(_Sat _Accum);

 
 int size_SatsF = sizeof(_Sat short _Fract);
 int size_SatF  = sizeof(_Sat _Fract);
 int size_SatlF = sizeof(_Sat long _Fract);
 int align_SatsF = __alignof(_Sat short _Fract);
 int align_SatF  = __alignof(_Sat _Fract);
 int align_SatlF = __alignof(_Sat long _Fract);
 
-// CHECK-NEXT: @size_SatsF   = {{.*}}global i{{[0-9]+}} 2
-// CHECK-NEXT: @size_SatF    = {{.*}}global i{{[0-9]+}} 4
-// CHECK-NEXT: @size_SatlF   = {{.*}}global i{{[0-9]+}} 8
-// CHECK-NEXT: @align_SatsF  = {{.*}}global i{{[0-9]+}} 2
-// CHECK-NEXT: @align_SatF   = {{.*}}global i{{[0-9]+}} 4
-// CHECK-NEXT: @align_SatlF  = {{.*}}global i{{[0-9]+}} 8
+// CHECK-NEXT: @size_SatsF   = {{.*}}global i{{[0-9]+}} 1
+// CHECK-NEXT: @size_SatF    = {{.*}}global i{{[0-9]+}} 2
+// CHECK-NEXT: @size_SatlF   = {{.*}}global i{{[0-9]+}} 4
+// CHECK-NEXT: @align_SatsF  = {{.*}}global i{{[0-9]+}} 1
+// CHECK-NEXT: @align_SatF   = {{.*}}global i{{[0-9]+}} 2
+// CHECK-NEXT: @align_SatlF  = {{.*}}global i{{[0-9]+}} 4

test/Frontend/fixed_point_declarations.c

+// RUN: %clang -ffixed-point -S -emit-llvm %s -o - | FileCheck %s
+// RUN: %clang -ffixed-point -S -emit-llvm %s -o - --target=x86_64-scei-ps4-ubuntu-fast | FileCheck %s
+
+// Primary fixed point types
+signed short _Accum   s_short_accum;  // CHECK-DAG: @s_short_accum  = {{.*}}global i16 0, align 2
+signed _Accum         s_accum;        // CHECK-DAG: @s_accum        = {{.*}}global i32 0, align 4
+signed long _Accum    s_long_accum;   // CHECK-DAG: @s_long_accum   = {{.*}}global i64 0, align 8
+unsigned short _Accum u_short_accum;  // CHECK-DAG: @u_short_accum  = {{.*}}global i16 0, align 2
+unsigned _Accum       u_accum;        // CHECK-DAG: @u_accum        = {{.*}}global i32 0, align 4
+unsigned long _Accum  u_long_accum;   // CHECK-DAG: @u_long_accum   = {{.*}}global i64 0, align 8
+signed short _Fract   s_short_fract;  // CHECK-DAG: @s_short_fract  = {{.*}}global i8  0, align 1
+signed _Fract         s_fract;        // CHECK-DAG: @s_fract        = {{.*}}global i16 0, align 2
+signed long _Fract    s_long_fract;   // CHECK-DAG: @s_long_fract   = {{.*}}global i32 0, align 4
+unsigned short _Fract u_short_fract;  // CHECK-DAG: @u_short_fract  = {{.*}}global i8  0, align 1
+unsigned _Fract       u_fract;        // CHECK-DAG: @u_fract        = {{.*}}global i16 0, align 2
+unsigned long _Fract  u_long_fract;   // CHECK-DAG: @u_long_fract   = {{.*}}global i32 0, align 4
+
+// Aliased
+short _Accum  short_accum;            // CHECK-DAG: @short_accum    = {{.*}}global i16 0, align 2
+_Accum        accum;                  // CHECK-DAG: @accum          = {{.*}}global i32 0, align 4
+long _Accum   long_accum;             // CHECK-DAG: @long_accum     = {{.*}}global i64 0, align 8
+short _Fract  short_fract;            // CHECK-DAG: @short_fract    = {{.*}}global i8  0, align 1
+_Fract        fract;                  // CHECK-DAG: @fract          = {{.*}}global i16 0, align 2
+long _Fract   long_fract;             // CHECK-DAG: @long_fract     = {{.*}}global i32 0, align 4
+
+// Primary saturated
+_Sat signed short _Accum   sat_s_short_accum;  // CHECK-DAG: @sat_s_short_accum  = {{.*}}global i16 0, align 2
+_Sat signed _Accum         sat_s_accum;        // CHECK-DAG: @sat_s_accum        = {{.*}}global i32 0, align 4
+_Sat signed long _Accum    sat_s_long_accum;   // CHECK-DAG: @sat_s_long_accum   = {{.*}}global i64 0, align 8
+_Sat unsigned short _Accum sat_u_short_accum;  // CHECK-DAG: @sat_u_short_accum  = {{.*}}global i16 0, align 2
+_Sat unsigned _Accum       sat_u_accum;        // CHECK-DAG: @sat_u_accum        = {{.*}}global i32 0, align 4
+_Sat unsigned long _Accum  sat_u_long_accum;   // CHECK-DAG: @sat_u_long_accum   = {{.*}}global i64 0, align 8
+_Sat signed short _Fract   sat_s_short_fract;  // CHECK-DAG: @sat_s_short_fract  = {{.*}}global i8  0, align 1
+_Sat signed _Fract         sat_s_fract;        // CHECK-DAG: @sat_s_fract        = {{.*}}global i16 0, align 2
+_Sat signed long _Fract    sat_s_long_fract;   // CHECK-DAG: @sat_s_long_fract   = {{.*}}global i32 0, align 4
+_Sat unsigned short _Fract sat_u_short_fract;  // CHECK-DAG: @sat_u_short_fract  = {{.*}}global i8  0, align 1
+_Sat unsigned _Fract       sat_u_fract;        // CHECK-DAG: @sat_u_fract        = {{.*}}global i16 0, align 2
+_Sat unsigned long _Fract  sat_u_long_fract;   // CHECK-DAG: @sat_u_long_fract   = {{.*}}global i32 0, align 4
+
+// Aliased saturated
+_Sat short _Accum  sat_short_accum;            // CHECK-DAG: @sat_short_accum    = {{.*}}global i16 0, align 2
+_Sat _Accum        sat_accum;                  // CHECK-DAG: @sat_accum          = {{.*}}global i32 0, align 4
+_Sat long _Accum   sat_long_accum;             // CHECK-DAG: @sat_long_accum     = {{.*}}global i64 0, align 8
+_Sat short _Fract  sat_short_fract;            // CHECK-DAG: @sat_short_fract    = {{.*}}global i8  0, align 1
+_Sat _Fract        sat_fract;                  // CHECK-DAG: @sat_fract          = {{.*}}global i16 0, align 2
+_Sat long _Fract   sat_long_fract;             // CHECK-DAG: @sat_long_fract     = {{.*}}global i32 0, align 4
+
+/* Fixed point literals */
+short _Accum  short_accum_literal = 2.5hk;    // CHECK-DAG: @short_accum_literal  = {{.*}}global i16 320, align 2
+_Accum        accum_literal       = 2.5k;     // CHECK-DAG: @accum_literal        = {{.*}}global i32 81920, align 4
+long _Accum   long_accum_literal  = 2.5lk;    // CHECK-DAG: @long_accum_literal   = {{.*}}global i64 5368709120, align 8
+short _Fract  short_fract_literal = 0.5hr;    // CHECK-DAG: @short_fract_literal  = {{.*}}global i8  64, align 1
+_Fract        fract_literal       = 0.5r;     // CHECK-DAG: @fract_literal        = {{.*}}global i16 16384, align 2
+long _Fract   long_fract_literal  = 0.5lr;    // CHECK-DAG: @long_fract_literal   = {{.*}}global i32 1073741824, align 4
+
+unsigned short _Accum  u_short_accum_literal = 2.5uhk;    // CHECK-DAG: @u_short_accum_literal  = {{.*}}global i16 640, align 2
+unsigned _Accum        u_accum_literal       = 2.5uk;     // CHECK-DAG: @u_accum_literal        = {{.*}}global i32 163840, align 4
+unsigned long _Accum   u_long_accum_literal  = 2.5ulk;    // CHECK-DAG: @u_long_accum_literal   = {{.*}}global i64 10737418240, align 8
+unsigned short _Fract  u_short_fract_literal = 0.5uhr;    // CHECK-DAG: @u_short_fract_literal  = {{.*}}global i8  -128, align 1
+unsigned _Fract        u_fract_literal       = 0.5ur;     // CHECK-DAG: @u_fract_literal        = {{.*}}global i16 -32768, align 2
+unsigned long _Fract   u_long_fract_literal  = 0.5ulr;    // CHECK-DAG: @u_long_fract_literal   = {{.*}}global i32 -2147483648, align 4
+
+// Max literal values
+short _Accum          short_accum_max   = 255.9999999999999999hk;         // CHECK-DAG: @short_accum_max   = {{.*}}global i16 32767, align 2
+_Accum                accum_max         = 65535.9999999999999999k;        // CHECK-DAG: @accum_max         = {{.*}}global i32 2147483647, align 4
+long _Accum           long_accum_max    = 4294967295.9999999999999999lk;  // CHECK-DAG: @long_accum_max    = {{.*}}global i64 9223372036854775807, align 8
+unsigned short _Accum u_short_accum_max = 255.9999999999999999uhk;        // CHECK-DAG: @u_short_accum_max = {{.*}}global i16 -1, align 2
+unsigned _Accum       u_accum_max       = 65535.9999999999999999uk;       // CHECK-DAG: @u_accum_max       = {{.*}}global i32 -1, align 4
+unsigned long _Accum  u_long_accum_max  = 4294967295.9999999999999999ulk; // CHECK-DAG: @u_long_accum_max  = {{.*}}global i64 -1, align 8
+
+short _Fract          short_fract_max   = 0.9999999999999999hr;           // CHECK-DAG: @short_fract_max   = {{.*}}global i8  127, align 1
+_Fract                fract_max         = 0.9999999999999999r;            // CHECK-DAG: @fract_max         = {{.*}}global i16 32767, align 2
+long _Fract           long_fract_max    = 0.9999999999999999lr;           // CHECK-DAG: @long_fract_max    = {{.*}}global i32 2147483647, align 4
+unsigned short _Fract u_short_fract_max = 0.9999999999999999uhr;          // CHECK-DAG: @u_short_fract_max = {{.*}}global i8  -1, align 1
+unsigned _Fract       u_fract_max       = 0.9999999999999999ur;           // CHECK-DAG: @u_fract_max       = {{.*}}global i16 -1, align 2
+unsigned long _Fract  u_long_fract_max  = 0.9999999999999999ulr;          // CHECK-DAG: @u_long_fract_max  = {{.*}}global i32 -1, align 4
+
+// Fracts may be exactly one but evaluate to the Fract max
+short _Fract          short_fract_one   = 1.0hr;    // CHECK-DAG: @short_fract_one    = {{.*}}global i8  127, align 1
+_Fract                fract_one         = 1.0r;     // CHECK-DAG: @fract_one          = {{.*}}global i16 32767, align 2
+long _Fract           long_fract_one    = 1.0lr;    // CHECK-DAG: @long_fract_one     = {{.*}}global i32 2147483647, align 4
+unsigned short _Fract u_short_fract_one = 1.0uhr;   // CHECK-DAG: @u_short_fract_one  = {{.*}}global i8  -1, align 1
+unsigned _Fract       u_fract_one       = 1.0ur;    // CHECK-DAG: @u_fract_one        = {{.*}}global i16 -1, align 2
+unsigned long _Fract  u_long_fract_one  = 1.0ulr;   // CHECK-DAG: @u_long_fract_one   = {{.*}}global i32 -1, align 4
+
+short _Fract          short_fract_exp_one   = 0.1e1hr;    // CHECK-DAG: @short_fract_exp_one    = {{.*}}global i8  127, align 1
+_Fract                fract_exp_one         = 0.1e1r;     // CHECK-DAG: @fract_exp_one          = {{.*}}global i16 32767, align 2
+long _Fract           long_fract_exp_one    = 0.1e1lr;    // CHECK-DAG: @long_fract_exp_one     = {{.*}}global i32 2147483647, align 4
+unsigned short _Fract u_short_fract_exp_one = 0.1e1uhr;   // CHECK-DAG: @u_short_fract_exp_one  = {{.*}}global i8  -1, align 1
+unsigned _Fract       u_fract_exp_one       = 0.1e1ur;    // CHECK-DAG: @u_fract_exp_one        = {{.*}}global i16 -1, align 2
+unsigned long _Fract  u_long_fract_exp_one  = 0.1e1ulr;   // CHECK-DAG: @u_long_fract_exp_one   = {{.*}}global i32 -1, align 4
+
+short _Fract          short_fract_hex_exp_one   = 0x0.8p1hr;    // CHECK-DAG: @short_fract_hex_exp_one    = {{.*}}global i8  127, align 1
+_Fract                fract_hex_exp_one         = 0x0.8p1r;     // CHECK-DAG: @fract_hex_exp_one          = {{.*}}global i16 32767, align 2
+long _Fract           long_fract_hex_exp_one    = 0x0.8p1lr;    // CHECK-DAG: @long_fract_hex_exp_one     = {{.*}}global i32 2147483647, align 4
+unsigned short _Fract u_short_fract_hex_exp_one = 0x0.8p1uhr;   // CHECK-DAG: @u_short_fract_hex_exp_one  = {{.*}}global i8  -1, align 1
+unsigned _Fract       u_fract_hex_exp_one       = 0x0.8p1ur;    // CHECK-DAG: @u_fract_hex_exp_one        = {{.*}}global i16 -1, align 2
+unsigned long _Fract  u_long_fract_hex_exp_one  = 0x0.8p1ulr;   // CHECK-DAG: @u_long_fract_hex_exp_one   = {{.*}}global i32 -1, align 4
+
+// Expsilon values
+short _Accum          short_accum_eps   = 0x1p-7hk;         // CHECK-DAG: @short_accum_eps   = {{.*}}global i16 1, align 2
+_Accum                accum_eps         = 0x1p-15k;         // CHECK-DAG: @accum_eps         = {{.*}}global i32 1, align 4
+long _Accum           long_accum_eps    = 0x1p-31lk;        // CHECK-DAG: @long_accum_eps    = {{.*}}global i64 1, align 8
+unsigned short _Accum u_short_accum_eps = 0x1p-8uhk;        // CHECK-DAG: @u_short_accum_eps = {{.*}}global i16 1, align 2
+unsigned _Accum       u_accum_eps       = 0x1p-16uk;        // CHECK-DAG: @u_accum_eps       = {{.*}}global i32 1, align 4
+unsigned long _Accum  u_long_accum_eps  = 0x1p-32ulk;       // CHECK-DAG: @u_long_accum_eps  = {{.*}}global i64 1, align 8
+
+short _Fract          short_fract_eps   = 0x1p-7hr;         // CHECK-DAG: @short_fract_eps   = {{.*}}global i8  1, align 1
+_Fract                fract_eps         = 0x1p-15r;         // CHECK-DAG: @fract_eps         = {{.*}}global i16 1, align 2
+long _Fract           long_fract_eps    = 0x1p-31lr;        // CHECK-DAG: @long_fract_eps    = {{.*}}global i32 1, align 4
+unsigned short _Fract u_short_fract_eps = 0x1p-8uhr;        // CHECK-DAG: @u_short_fract_eps = {{.*}}global i8  1, align 1
+unsigned _Fract       u_fract_eps       = 0x1p-16ur;        // CHECK-DAG: @u_fract_eps       = {{.*}}global i16 1, align 2
+unsigned long _Fract  u_long_fract_eps  = 0x1p-32ulr;       // CHECK-DAG: @u_long_fract_eps  = {{.*}}global i32 1, align 4

test/Frontend/fixed_point_errors.c

 /* Bad saturated combinations */
 _Sat float f;             // expected-error{{'_Sat' specifier is only valid on '_Fract' or '_Accum', not 'float'}}
 _Sat double d;            // expected-error{{'_Sat' specifier is only valid on '_Fract' or '_Accum', not 'double'}}
 _Sat _Bool b;             // expected-error{{'_Sat' specifier is only valid on '_Fract' or '_Accum', not '_Bool'}}
 _Sat char c;              // expected-error{{'_Sat' specifier is only valid on '_Fract' or '_Accum', not 'char'}}
 _Sat int i;               // expected-error{{'_Sat' specifier is only valid on '_Fract' or '_Accum', not 'int'}}
 _Sat _Sat _Fract fract;   // expected-warning{{duplicate '_Sat' declaration specifier}}
 
+
+/* Literals that cannot fit into types */
+signed short _Accum s_short_accum = 256.0hk;            // expected-error{{this value is too large for this fixed point type}}
+unsigned short _Accum u_short_accum = 256.0uhk;         // expected-error{{this value is too large for this fixed point type}}
+signed _Accum s_accum = 65536.0k;                       // expected-error{{this value is too large for this fixed point type}}
+unsigned _Accum u_accum = 65536.0uk;                    // expected-error{{this value is too large for this fixed point type}}
+signed long _Accum s_long_accum = 4294967296.0lk;       // expected-error{{this value is too large for this fixed point type}}
+unsigned long _Accum u_long_accum = 4294967296.0ulk;    // expected-error{{this value is too large for this fixed point type}}
+
+// Large values from decimal exponents
+short _Accum          short_accum_exp   = 2.56e2hk;           // expected-error{{this value is too large for this fixed point type}}
+_Accum                accum_exp         = 6.5536e4k;          // expected-error{{this value is too large for this fixed point type}}
+long _Accum           long_accum_exp    = 4.294967296e9lk;    // expected-error{{this value is too large for this fixed point type}}
+unsigned short _Accum u_short_accum_exp = 2.56e2uhk;          // expected-error{{this value is too large for this fixed point type}}
+unsigned _Accum       u_accum_exp       = 6.5536e4uk;         // expected-error{{this value is too large for this fixed point type}}
+unsigned long _Accum  u_long_accum_exp  = 4.294967296e9ulk;   // expected-error{{this value is too large for this fixed point type}}
+
+// Large value from hexidecimal exponents
+short _Accum          short_accum_hex_exp   = 0x1p8hk;        // expected-error{{this value is too large for this fixed point type}}
+_Accum                accum_hex_exp         = 0x1p16k;        // expected-error{{this value is too large for this fixed point type}}
+long _Accum           long_accum_hex_exp    = 0x1p32lk;       // expected-error{{this value is too large for this fixed point type}}
+unsigned short _Accum u_short_accum_hex_exp = 0x1p8uhk;       // expected-error{{this value is too large for this fixed point type}}
+unsigned _Accum       u_accum_hex_exp       = 0x1p16uk;       // expected-error{{this value is too large for this fixed point type}}
+unsigned long _Accum  u_long_accum_hex_exp  = 0x1p32ulk;      // expected-error{{this value is too large for this fixed point type}}
+
+// Very large exponent
+_Accum x = 1e1000000000000000000000000000000000k;   // expected-error{{this value is too large for this fixed point type}}
+
+/* Although _Fract's cannot equal 1, _Fract literals written as 1 are allowed
+ * and the underlying value represents the max value for that _Fract type. */
+short _Fract          short_fract_above_1    = 1.1hr;   // expected-error{{this value is too large for this fixed point type}}
+_Fract                fract_above_1          = 1.1r;    // expected-error{{this value is too large for this fixed point type}}
+long _Fract           long_fract_above_1     = 1.1lr;   // expected-error{{this value is too large for this fixed point type}}
+unsigned short _Fract u_short_fract_above_1  = 1.1uhr;  // expected-error{{this value is too large for this fixed point type}}
+unsigned _Fract       u_fract_above_1        = 1.1ur;   // expected-error{{this value is too large for this fixed point type}}
+unsigned long _Fract  u_long_fract_above_1   = 1.1ulr;  // expected-error{{this value is too large for this fixed point type}}
+
+short _Fract          short_fract_hex_exp   = 0x0.fp1hr;      // expected-error{{this value is too large for this fixed point type}}
+_Fract                fract_hex_exp         = 0x0.fp1r;       // expected-error{{this value is too large for this fixed point type}}
+long _Fract           long_fract_hex_exp    = 0x0.fp1lr;      // expected-error{{this value is too large for this fixed point type}}
+unsigned short _Fract u_short_fract_hex_exp = 0x0.fp1uhr;     // expected-error{{this value is too large for this fixed point type}}
+unsigned _Fract       u_fract_hex_exp       = 0x0.fp1ur;      // expected-error{{this value is too large for this fixed point type}}
+unsigned long _Fract  u_long_fract_hex_exp  = 0x0.fp1ulr;     // expected-error{{this value is too large for this fixed point type}}
+
 /* Do not allow typedef to be used with typedef'd types */
 typedef short _Fract shortfract_t;
 typedef short _Accum shortaccum_t;
 typedef _Fract fract_t;
 typedef _Accum accum_t;
 typedef long _Fract longfract_t;
 typedef long _Accum longaccum_t;
 _Sat shortfract_t td_sat_short_fract;       // expected-error{{'_Sat' specifier is only valid on '_Fract' or '_Accum', not 'type-name'}}
 _Sat shortaccum_t td_sat_short_accum;       // expected-error{{'_Sat' specifier is only valid on '_Fract' or '_Accum', not 'type-name'}}
 _Sat fract_t td_sat_fract;                  // expected-error{{'_Sat' specifier is only valid on '_Fract' or '_Accum', not 'type-name'}}
 _Sat accum_t td_sat_accum;                  // expected-error{{'_Sat' specifier is only valid on '_Fract' or '_Accum', not 'type-name'}}
 _Sat longfract_t td_sat_long_fract;         // expected-error{{'_Sat' specifier is only valid on '_Fract' or '_Accum', not 'type-name'}}
 _Sat longaccum_t td_sat_long_accum;         // expected-error{{'_Sat' specifier is only valid on '_Fract' or '_Accum', not 'type-name'}}
+
+/* Bad suffixes  */
+_Accum fk = 1.0fk;    // expected-error{{invalid suffix 'fk' on integer constant}}
+_Accum kk = 1.0kk;    // expected-error{{invalid suffix 'kk' on integer constant}}
+_Accum rk = 1.0rk;    // expected-error{{invalid suffix 'rk' on integer constant}}
+_Accum rk = 1.0rr;    // expected-error{{invalid suffix 'rr' on integer constant}}
+_Accum qk = 1.0qr;    // expected-error{{invalid suffix 'qr' on integer constant}}
+
+/* Using wrong exponent notation */
+_Accum dec_with_hex_exp1 = 0.1p10k;    // expected-error{{invalid suffix 'p10k' on integer constant}}
+_Accum dec_with_hex_exp2 = 0.1P10k;    // expected-error{{invalid suffix 'P10k' on integer constant}}
+_Accum hex_with_dex_exp1 = 0x0.1e10k;  // expected-error{{hexadecimal floating constant requires an exponent}}
+_Accum hex_with_dex_exp2 = 0x0.1E10k;  // expected-error{{hexadecimal floating constant requires an exponent}}

test/Frontend/fixed_point_same_fbits.c

+// RUN: %clang -ffixed-point -S -emit-llvm -o - %s | FileCheck %s -check-prefix=DEFAULT
+// RUN: %clang -ffixed-point -fsame-fbits -S -emit-llvm -o - %s | FileCheck %s -check-prefix=SAME
+
+/* The scale for unsigned fixed point types should be the same as that of signed
+ * fixed point types when -fsame-fbits is enabled. */
+
+void func() {
+  unsigned short _Accum u_short_accum = 0.5uhk;
+  unsigned _Accum u_accum = 0.5uk;
+  unsigned long _Accum u_long_accum = 0.5ulk;
+  unsigned short _Fract u_short_fract = 0.5uhr;
+  unsigned _Fract u_fract = 0.5ur;
+  unsigned long _Fract u_long_fract = 0.5ulr;
+
+// DEFAULT: store i16 128, i16* %u_short_accum, align 2
+// DEFAULT: store i32 32768, i32* %u_accum, align 4
+// DEFAULT: store i64 2147483648, i64* %u_long_accum, align 8
+// DEFAULT: store i8  -128, i8* %u_short_fract, align 1
+// DEFAULT: store i16 -32768, i16* %u_fract, align 2
+// DEFAULT: store i32 -2147483648, i32* %u_long_fract, align 4
+
+// SAME: store i16 64, i16* %u_short_accum, align 2
+// SAME: store i32 16384, i32* %u_accum, align 4
+// SAME: store i64 1073741824, i64* %u_long_accum, align 8
+// SAME: store i8  64, i8* %u_short_fract, align 1
+// SAME: store i16 16384, i16* %u_fract, align 2
+// SAME: store i32 1073741824, i32* %u_long_fract, align 4
+}

tools/libclang/CIndex.cpp

CXString clang_getCursorKindSpelling(enum CXCursorKind Kind) {
   case CXCursor_LabelRef:
     return cxstring::createRef("LabelRef");
   case CXCursor_OverloadedDeclRef:
     return cxstring::createRef("OverloadedDeclRef");
   case CXCursor_VariableRef:
     return cxstring::createRef("VariableRef");
   case CXCursor_IntegerLiteral:
       return cxstring::createRef("IntegerLiteral");
+  case CXCursor_FixedPointLiteral:
+      return cxstring::createRef("FixedPointLiteral");
   case CXCursor_FloatingLiteral:
       return cxstring::createRef("FloatingLiteral");
   case CXCursor_ImaginaryLiteral:
       return cxstring::createRef("ImaginaryLiteral");
   case CXCursor_StringLiteral:
       return cxstring::createRef("StringLiteral");
   case CXCursor_CharacterLiteral:
       return cxstring::createRef("CharacterLiteral");

tools/libclang/CXCursor.cpp

CXCursor cxcursor::MakeCXCursor(const Stmt *S, const Decl *Parent,
   case Stmt::CapturedStmtClass:
     K = CXCursor_UnexposedStmt;
     break;
 
   case Stmt::IntegerLiteralClass:
     K = CXCursor_IntegerLiteral;
     break;
 
+  case Stmt::FixedPointLiteralClass:
+    K = CXCursor_FixedPointLiteral;
+    break;
+
   case Stmt::FloatingLiteralClass:
     K = CXCursor_FloatingLiteral;
     break;
 
   case Stmt::ImaginaryLiteralClass:
     K = CXCursor_ImaginaryLiteral;
     break;