Thanks to Eli Friedman, who suggested the reason these tests failed on a few
buildbots yet works fine locally is because non-assert builds don't emit value
labels.
llvm-svn: 322514
RISCVABIInfo is implemented in terms of XLen, supporting both RV32 and RV64.
Unfortunately we need to count argument registers in the frontend in order to
determine when to emit signext and zeroext attributes. Integer scalars are
extended according to their type up to 32-bits and then sign-extended to XLen
when passed in registers, but are anyext when passed on the stack. This patch
only implements the base integer (soft float) ABIs.
For more information on the RISC-V ABI, see [the ABI
doc](https://github.com/riscv/riscv-elf-psabi-doc/blob/master/riscv-elf.md),
my [golden model](https://github.com/lowRISC/riscv-calling-conv-model), and
the [LLVM RISC-V calling convention
patch](https://reviews.llvm.org/D39898#2d1595b4) (specifically the comment
documenting frontend expectations).
Differential Revision: https://reviews.llvm.org/D40023
llvm-svn: 322494
Summary:
kunpck intrinsics were removed in favor of native IR a few months ago. The implementation lowers them as by operation on the integer types passed to the intrinsic and then just shifting, masking, and oring them together. A special X86 DAG combine was added to recognize this patter and turn it into a concat_vector operation.
I think it makes more sense to keep the IR implementation closer to vector operations on vXi1. Given that we expect these builtins to be used around other builtins that operate on k-registers which we try to represent in IR with vXi1. InstCombine should be able to get rid of the bitcasts between integers and vXi1 leaving only the vector operations.
Reviewers: RKSimon, spatel, zvi, jina.nahias
Reviewed By: RKSimon
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D42016
llvm-svn: 322461
GCOV in the old pass manager also strips debug info (if debug info is
disabled/only produced for profiling anyway) after the GCOV pass runs.
I think the strip pass hasn't been ported to the new pass manager, so it
might take me a little while to wire that up.
llvm-svn: 322126
Cf-protection is a target independent flag that instructs the back-end to instrument control flow mechanisms like: Branch, Return, etc.
For example in X86 this flag will be used to instrument Indirect Branch Tracking instructions.
Differential Revision: https://reviews.llvm.org/D40478
Change-Id: I5126e766c0e6b84118cae0ee8a20fe78cc373dea
llvm-svn: 322063
GCC's attribute 'target', in addition to being an optimization hint,
also allows function multiversioning. We currently have the former
implemented, this is the latter's implementation.
This works by enabling functions with the same name/signature to coexist,
so that they can all be emitted. Multiversion state is stored in the
FunctionDecl itself, and SemaDecl manages the definitions.
Note that it ends up having to permit redefinition of functions so
that they can all be emitted. Additionally, all versions of the function
must be emitted, so this also manages that.
Note that this includes some additional rules that GCC does not, since
defining something as a MultiVersion function after a usage has been made illegal.
The only 'history rewriting' that happens is if a function is emitted before
it has been converted to a multiversion'ed function, at which point its name
needs to be changed.
Function templates and virtual functions are NOT yet supported (not supported
in GCC either).
Additionally, constructors/destructors are disallowed, but the former is
planned.
llvm-svn: 322028
Adds the -fstack-size-section flag to enable the .stack_sizes section. The flag defaults to on for the PS4 triple.
Differential Revision: https://reviews.llvm.org/D40712
llvm-svn: 321992
These just overloads for _Float128. They're supported by GCC 7 and used
by glibc. APFloat support is already there so just add the overloads.
__builtin_copysignf128
__builtin_fabsf128
__builtin_huge_valf128
__builtin_inff128
__builtin_nanf128
__builtin_nansf128
This is the same support that GCC has, according to the documentation,
but limited to _Float128.
llvm-svn: 321948
r320902 fixed the IRGen for some types of checked multiplications. It
did not handle unsigned overflow correctly in the case where the signed
operand is negative (PR35750).
Eli pointed out that on overflow, the result must be equal to the unique
value that is equivalent to the mathematically-correct result modulo two
raised to the k power, where k is the number of bits in the result type.
This patch fixes the specialized IRGen from r320902 accordingly.
Testing: Apart from check-clang, I modified the test harness from
r320902 to validate the results of all multiplications -- not just the
ones which don't overflow:
https://gist.github.com/vedantk/3eb9c88f82e5c32f2e590555b4af5081
llvm.org/PR35750, rdar://34963321
Differential Revision: https://reviews.llvm.org/D41717
llvm-svn: 321771
Summary:
The C++ Itanium ABI says:
No cookie is required if the new operator being used is ::operator new[](size_t, void*).
We should only avoid poisoning the cookie if we're calling this
operator, not others. This is dealt with before the call to
InitializeArrayCookie.
Reviewers: rjmccall, kcc, rsmith
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D41301
llvm-svn: 321645
added vbmi2 feature recognition
added intrinsics support for vbmi2 instructions
_mm[128,256,512]_mask[z]_compress_epi[16,32]
_mm[128,256,512]_mask_compressstoreu_epi[16,32]
_mm[128,256,512]_mask[z]_expand_epi[16,32]
_mm[128,256,512]_mask[z]_expandloadu_epi[16,32]
_mm[128,256,512]_mask[z]_sh[l,r]di_epi[16,32,64]
_mm[128,256,512]_mask_sh[l,r]dv_epi[16,32,64]
matching a similar work on the backend (D40206)
Differential Revision: https://reviews.llvm.org/D41557
llvm-svn: 321487
added vpclmulqdq feature recognition
added intrinsics support for vpclmulqdq instructions
_mm256_clmulepi64_epi128
_mm512_clmulepi64_epi128
matching a similar work on the backend (D40101)
Differential Revision: https://reviews.llvm.org/D41573
llvm-svn: 321480
added vaes feature recognition
added intrinsics support for vaes instructions, matching a similar work on the backend (D40078)
_mm256_aesenc_epi128
_mm512_aesenc_epi128
_mm256_aesenclast_epi128
_mm512_aesenclast_epi128
_mm256_aesdec_epi128
_mm512_aesdec_epi128
_mm256_aesdeclast_epi128
_mm512_aesdeclast_epi128
llvm-svn: 321474
Now that in the new TBAA format we allow access types to be of
any object types, including aggregate ones, it becomes critical
to specify types of all sub-objects such aggregates comprise as
their members. In order to meet this requirement, this patch
enables generation of field descriptors for members of array
types.
Differential Revision: https://reviews.llvm.org/D41399
llvm-svn: 321352
Now that the MDBuilder helpers generating TBAA type and access
descriptors in the new format are in place, we can teach clang to
use them when requested.
Differential Revision: https://reviews.llvm.org/D41394
llvm-svn: 321351
When a function taking transparent union is declared as taking one of
union members earlier in the translation unit, clang would hit an
"Invalid cast" assertion during EmitFunctionProlog. This case
corresponds to function f1 in test/CodeGen/transparent-union-redecl.c.
We decided to cast i32 to union because after merging function
declarations function parameter type becomes int,
CGFunctionInfo::ArgInfo type matches with ABIArgInfo type, so we decide
it is a trivial case. But these types should also be castable to
parameter declaration type which is not the case here.
The fix is in checking for the trivial case if ABIArgInfo type matches with
parameter declaration type. It exposed inconsistency that we check
hasScalarEvaluationKind for different types in EmitParmDecl and
EmitFunctionProlog, and comment says they should match.
Additional tests in Sema/transparent-union.c capture current behavior and make
sure there are no regressions.
rdar://problem/34949329
Reviewers: rjmccall, rafael
Reviewed By: rjmccall
Subscribers: aemerson, cfe-commits, kristof.beyls
Differential Revision: https://reviews.llvm.org/D41311
llvm-svn: 321296
Diagnose 'unreachable' UB when a noreturn function returns.
1. Insert a check at the end of functions marked noreturn.
2. A decl may be marked noreturn in the caller TU, but not marked in
the TU where it's defined. To diagnose this scenario, strip away the
noreturn attribute on the callee and insert check after calls to it.
Testing: check-clang, check-ubsan, check-ubsan-minimal, D40700
rdar://33660464
Differential Revision: https://reviews.llvm.org/D40698
llvm-svn: 321231
Summary: Plant an inline version of "((ac+bd)/(cc+dd)) + i((bc-ad)/(cc+dd))" instead.
Patch by Paul Walker.
Reviewed By: hfinkel
Differential Revision: https://reviews.llvm.org/D40299
llvm-svn: 321183
There are 2 parts to getting the -fassociative-math command-line flag translated to LLVM FMF:
1. In the driver/frontend, we accept the flag and its 'no' inverse and deal with the
interactions with other flags like -ffast-math -fno-signed-zeros -fno-trapping-math.
This was mostly already done - we just need to translate the flag as a codegen option.
The test file is complicated because there are many potential combinations of flags here.
Note that we are matching gcc's behavior that requires 'nsz' and no-trapping-math.
2. In codegen, we map the codegen option to FMF in the IR builder. This is simple code and
corresponding test.
For the motivating example from PR27372:
float foo(float a, float x) { return ((a + x) - x); }
$ ./clang -O2 27372.c -S -o - -ffast-math -fno-associative-math -emit-llvm | egrep 'fadd|fsub'
%add = fadd nnan ninf nsz arcp contract float %0, %1
%sub = fsub nnan ninf nsz arcp contract float %add, %2
So 'reassoc' is off as expected (and so is the new 'afn' but that's a different patch).
This case now works as expected end-to-end although the underlying logic is still wrong:
$ ./clang -O2 27372.c -S -o - -ffast-math -fno-associative-math | grep xmm
addss %xmm1, %xmm0
subss %xmm1, %xmm0
We're not done because the case where 'reassoc' is set is ignored by optimizer passes. Example:
$ ./clang -O2 27372.c -S -o - -fassociative-math -fno-signed-zeros -fno-trapping-math -emit-llvm | grep fadd
%add = fadd reassoc float %0, %1
$ ./clang -O2 27372.c -S -o - -fassociative-math -fno-signed-zeros -fno-trapping-math | grep xmm
addss %xmm1, %xmm0
subss %xmm1, %xmm0
Differential Revision: https://reviews.llvm.org/D39812
llvm-svn: 320920
This patch introduces a specialized way to lower overflow-checked
multiplications with mixed-sign operands. This fixes link failures and
ICEs on code like this:
void mul(int64_t a, uint64_t b) {
int64_t res;
__builtin_mul_overflow(a, b, &res);
}
The generic checked-binop irgen would use a 65-bit multiplication
intrinsic here, which requires runtime support for _muloti4 (128-bit
multiplication), and therefore fails to link on i386. To get an ICE
on x86_64, change the example to use __int128_t / __uint128_t.
Adding runtime and backend support for 65-bit or 129-bit checked
multiplication on all of our supported targets is infeasible.
This patch solves the problem by using simpler, specialized irgen for
the mixed-sign case.
llvm.org/PR34920, rdar://34963321
Testing: Apart from check-clang, I compared the output from this fairly
comprehensive test driver using unpatched & patched clangs:
https://gist.github.com/vedantk/3eb9c88f82e5c32f2e590555b4af5081
Differential Revision: https://reviews.llvm.org/D41149
llvm-svn: 320902
Summary:
InterlockedCompareExchange128 is a bit more complicated than the other
InterlockedCompareExchange functions, so it requires a bit more work. It
doesn't directly refer to 128bit ints, instead it takes pointers to
64bit ints for Destination and ComparandResult, and exchange is taken as
two 64bit ints (high & low). The previous value is written to
ComparandResult, and success is returned. This implementation does the
following in order to produce a cmpxchg instruction:
1. Cast everything to 128bit ints or int pointers, and glues together
the Exchange values
2. Reads from CompareandResult to get the comparand
3. Calls cmpxchg volatile (on X86 this will produce a lock cmpxchg16b
instruction)
1. Result 0 (previous value) is written back to ComparandResult
2. Result 1 (success bool) is zext'ed to a uchar and returned
Resolves bug https://llvm.org/PR35251
Patch by Colden Cullen!
Reviewers: rnk, agutowski
Reviewed By: rnk
Subscribers: majnemer, cfe-commits
Differential Revision: https://reviews.llvm.org/D41032
llvm-svn: 320730
This adds a new command line option -mprefer-vector-width to specify a preferred vector width for the vectorizers. Valid values are 'none' and unsigned integers. The driver will check that it meets those constraints. Specific supported integers will be managed by the targets in the backend.
Clang will take the value and add it as a new function attribute during CodeGen.
This represents the alternate direction proposed by Sanjay in this RFC: http://lists.llvm.org/pipermail/llvm-dev/2017-November/118734.html
The syntax here matches gcc, though gcc treats it as an x86 specific command line argument. gcc only allows values of 128, 256, and 512. I'm not having clang check any values.
Differential Revision: https://reviews.llvm.org/D40230
llvm-svn: 320419
This commit fixes a bug in IRGen where it generates completely broken
code for __fp16 vectors on X86. For example when the following code is
compiled:
half4 hv0, hv1, hv2; // these are vectors of __fp16.
void foo221() {
hv0 = hv1 + hv2;
}
clang generates the following IR, in which two i16 vectors are added:
@hv1 = common global <4 x i16> zeroinitializer, align 8
@hv2 = common global <4 x i16> zeroinitializer, align 8
@hv0 = common global <4 x i16> zeroinitializer, align 8
define void @foo221() {
%0 = load <4 x i16>, <4 x i16>* @hv1, align 8
%1 = load <4 x i16>, <4 x i16>* @hv2, align 8
%add = add <4 x i16> %0, %1
store <4 x i16> %add, <4 x i16>* @hv0, align 8
ret void
}
To fix the bug, this commit uses the code committed in r314056, which
modified clang to promote and truncate __fp16 vectors to and from float
vectors in the AST. It also fixes another IRGen bug where a short value
is assigned to an __fp16 variable without any integer-to-floating-point
conversion, as shown in the following example:
__fp16 a;
short b;
void foo1() {
a = b;
}
@b = common global i16 0, align 2
@a = common global i16 0, align 2
define void @foo1() #0 {
%0 = load i16, i16* @b, align 2
store i16 %0, i16* @a, align 2
ret void
}
rdar://problem/20625184
Differential Revision: https://reviews.llvm.org/D40112
llvm-svn: 320215
This is a follow-up to r320128. Eli pointed out that there is some gray
area in the language standard about whether the constant size is exact,
or a lower bound.
https://reviews.llvm.org/D40940
llvm-svn: 320185
This patch, together with a matching llvm patch (https://reviews.llvm.org/D39720), implements the lowering of X86 kunpack intrinsics to IR.
Differential Revision: https://reviews.llvm.org/D39719
Change-Id: Id5d3cb394ad33b98be79a6783d1d15569e2b798d
llvm-svn: 319777
There are 20 LLVM math intrinsics that correspond to mathlib calls according to the LangRef:
http://llvm.org/docs/LangRef.html#standard-c-library-intrinsics
We were only converting 3 mathlib calls (sqrt, fma, pow) and 12 builtin calls (ceil, copysign,
fabs, floor, fma, fmax, fmin, nearbyint, pow, rint, round, trunc) to their intrinsic-equivalents.
This patch pulls the transforms together and handles all 20 cases. The switch is guarded by a
check for const-ness to make sure we're not doing the transform if errno could possibly be set by
the libcall or builtin.
Differential Revision: https://reviews.llvm.org/D40044
llvm-svn: 319593
George Burgess IV