maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/clang/test/CodeGen/builtins.c

847 lines
33 KiB
C
Raw Normal View History

Update tests to use %clang_cc1 instead of 'clang-cc' or 'clang -cc1'. - This is designed to make it obvious that %clang_cc1 is a "test variable" which is substituted. It is '%clang_cc1' instead of '%clang -cc1' because it can be useful to redefine what gets run as 'clang -cc1' (for example, to set a default target). llvm-svn: 91446
2009-12-16 04:14:24 +08:00
// RUN: %clang_cc1 -emit-llvm -o %t %s
Add two test cases for builtins (mostly related to object size builtins). llvm-svn: 55736
2008-09-04 05:17:21 +08:00
// RUN: not grep __builtin %t
Support constant-evaluation of __builtin_nans* as well as the correct constant evaluation of __builtin_nan*. Most of the work to make this work is in LLVM. Fixes <rdar://problem/7696712> and part of PR 5255. llvm-svn: 97383
2010-02-28 21:00:19 +08:00
// RUN: %clang_cc1 %s -emit-llvm -o - -triple x86_64-darwin-apple | FileCheck %s
Add two test cases for builtins (mostly related to object size builtins). llvm-svn: 55736
2008-09-04 05:17:21 +08:00
Don't #include <stdio.h> when tests don't need it, or use clang instead of clang-cc when they do. llvm-svn: 89070
2009-11-17 16:57:36 +08:00
int printf(const char *, ...);
Add two test cases for builtins (mostly related to object size builtins). llvm-svn: 55736
2008-09-04 05:17:21 +08:00
void p(char *str, int x) {
printf("%s: %d\n", str, x);
}
void q(char *str, double x) {
printf("%s: %f\n", str, x);
}
Support constant-evaluation of __builtin_nans* as well as the correct constant evaluation of __builtin_nan*. Most of the work to make this work is in LLVM. Fixes <rdar://problem/7696712> and part of PR 5255. llvm-svn: 97383
2010-02-28 21:00:19 +08:00
void r(char *str, void *ptr) {
printf("%s: %p\n", str, ptr);
}
int random(void);
Add two test cases for builtins (mostly related to object size builtins). llvm-svn: 55736
2008-09-04 05:17:21 +08:00
int main() {
int N = random();
#define P(n,args) p(#n #args, __builtin_##n args)
#define Q(n,args) q(#n #args, __builtin_##n args)
Support constant-evaluation of __builtin_nans* as well as the correct constant evaluation of __builtin_nan*. Most of the work to make this work is in LLVM. Fixes <rdar://problem/7696712> and part of PR 5255. llvm-svn: 97383
2010-02-28 21:00:19 +08:00
#define R(n,args) r(#n #args, __builtin_##n args)
Add two test cases for builtins (mostly related to object size builtins). llvm-svn: 55736
2008-09-04 05:17:21 +08:00
#define V(n,args) p(#n #args, (__builtin_##n args, 0))
P(types_compatible_p, (int, float));
P(choose_expr, (0, 10, 20));
P(constant_p, (sizeof(10)));
P(expect, (N == 12, 0));
V(prefetch, (&N));
V(prefetch, (&N, 1));
V(prefetch, (&N, 1, 0));
// Numeric Constants
Q(huge_val, ());
Q(huge_valf, ());
Q(huge_vall, ());
Q(inf, ());
Q(inff, ());
Q(infl, ());
Enable basic testing of __builtin_fpclassify. llvm-svn: 105937
2010-06-14 18:41:45 +08:00
P(fpclassify, (0, 1, 2, 3, 4, 1.0));
P(fpclassify, (0, 1, 2, 3, 4, 1.0f));
P(fpclassify, (0, 1, 2, 3, 4, 1.0l));
Add two test cases for builtins (mostly related to object size builtins). llvm-svn: 55736
2008-09-04 05:17:21 +08:00
Add some builtins to codegen test case. llvm-svn: 57101
2008-10-05 14:34:45 +08:00
Q(nan, (""));
Q(nanf, (""));
Q(nanl, (""));
Q(nans, (""));
Q(nan, ("10"));
Q(nanf, ("10"));
Q(nanl, ("10"));
Q(nans, ("10"));
P(isgreater, (1., 2.));
P(isgreaterequal, (1., 2.));
P(isless, (1., 2.));
P(islessequal, (1., 2.));
P(islessgreater, (1., 2.));
P(isunordered, (1., 2.));
Add two test cases for builtins (mostly related to object size builtins). llvm-svn: 55736
2008-09-04 05:17:21 +08:00
[Modules] Implement __builtin_isinf_sign in Clang. Somehow, we never managed to implement this fully. We could constant fold it like crazy, including constant folding complex arguments, etc. But if you actually needed to generate code for it, error. I've implemented it using the somewhat obvious lowering. Happy for suggestions on a more clever way to lower this. Now, what you might ask does this have to do with modules? Fun story. So it turns out that libstdc++ actually uses __builtin_isinf_sign to implement std::isinf when in C++98 mode, but only inside of a template. So if we're lucky, and we never instantiate that, everything is good. But once we try to instantiate that template function, we need this builtin. All of my customers at least are using C++11 and so they never hit this code path. But what does that have to do with modules? Fun story. So it turns out that with modules we actually observe a bunch of bugs in libstdc++ where their <cmath> header clobbers things exposed by <math.h>. To fix these, we have to provide global function definitions to replace the macros that C99 would have used. And it turns out that ::isinf needs to be implemented using the exact semantics used by the C++98 variant of std::isinf. And so I started to fix this bug in libstdc++ and ceased to be able to compile libstdc++ with Clang. The yaks are legion. llvm-svn: 232778
2015-03-20 06:39:51 +08:00
P(isinf, (1.));
P(isinf_sign, (1.));
PR4836, part 2: CodeGen for __builtin_isnan. llvm-svn: 80655
2009-09-01 12:19:44 +08:00
P(isnan, (1.));
Add two test cases for builtins (mostly related to object size builtins). llvm-svn: 55736
2008-09-04 05:17:21 +08:00
// Bitwise & Numeric Functions
Add some builtins to codegen test case. llvm-svn: 57101
2008-10-05 14:34:45 +08:00
P(abs, (N));
Add two test cases for builtins (mostly related to object size builtins). llvm-svn: 55736
2008-09-04 05:17:21 +08:00
P(clz, (N));
P(clzl, (N));
P(clzll, (N));
P(ctz, (N));
P(ctzl, (N));
P(ctzll, (N));
P(ffs, (N));
P(ffsl, (N));
P(ffsll, (N));
P(parity, (N));
P(parityl, (N));
P(parityll, (N));
P(popcount, (N));
P(popcountl, (N));
P(popcountll, (N));
Q(powi, (1.2f, N));
Q(powif, (1.2f, N));
Q(powil, (1.2f, N));
Add some stdlib builtins llvm-svn: 65115
2009-02-20 14:36:40 +08:00
// Lib functions
Add two test cases for builtins (mostly related to object size builtins). llvm-svn: 55736
2008-09-04 05:17:21 +08:00
int a, b, n = random(); // Avoid optimizing out.
char s0[10], s1[] = "Hello";
Add some stdlib builtins llvm-svn: 65115
2009-02-20 14:36:40 +08:00
V(strcat, (s0, s1));
Add more builtin definitions, including strcmp. - PR3964. llvm-svn: 68710
2009-04-10 00:42:50 +08:00
V(strcmp, (s0, s1));
Add some stdlib builtins llvm-svn: 65115
2009-02-20 14:36:40 +08:00
V(strncat, (s0, s1, n));
V(strchr, (s0, s1[0]));
V(strrchr, (s0, s1[0]));
V(strcpy, (s0, s1));
V(strncpy, (s0, s1, n));
// Object size checking
Add two test cases for builtins (mostly related to object size builtins). llvm-svn: 55736
2008-09-04 05:17:21 +08:00
V(__memset_chk, (s0, 0, sizeof s0, n));
V(__memcpy_chk, (s0, s1, sizeof s0, n));
V(__memmove_chk, (s0, s1, sizeof s0, n));
V(__mempcpy_chk, (s0, s1, sizeof s0, n));
V(__strncpy_chk, (s0, s1, sizeof s0, n));
V(__strcpy_chk, (s0, s1, n));
s0[0] = 0;
V(__strcat_chk, (s0, s1, n));
P(object_size, (s0, 0));
P(object_size, (s0, 1));
P(object_size, (s0, 2));
P(object_size, (s0, 3));
// Whatever
Expose __builtin_bswap16. GCC has always supported this on PowerPC and 4.8 supports it on all platforms, so it's a good idea to expose it in clang too. LLVM supports this on all targets. llvm-svn: 165362
2012-10-06 22:42:22 +08:00
P(bswap16, (N));
Add two test cases for builtins (mostly related to object size builtins). llvm-svn: 55736
2008-09-04 05:17:21 +08:00
P(bswap32, (N));
P(bswap64, (N));
Add builtins for bitreverse intrinsic Follow the naming convention that bswap uses since it's a similar sort of operation. llvm-svn: 259671
2016-02-04 01:49:38 +08:00
Add missing __builtin_bitreverse8 Also add documentation for bitreverse builtins llvm-svn: 264203
2016-03-24 06:14:43 +08:00
// CHECK: @llvm.bitreverse.i8
Add builtins for bitreverse intrinsic Follow the naming convention that bswap uses since it's a similar sort of operation. llvm-svn: 259671
2016-02-04 01:49:38 +08:00
// CHECK: @llvm.bitreverse.i16
// CHECK: @llvm.bitreverse.i32
// CHECK: @llvm.bitreverse.i64
Add missing __builtin_bitreverse8 Also add documentation for bitreverse builtins llvm-svn: 264203
2016-03-24 06:14:43 +08:00
P(bitreverse8, (N));
Add builtins for bitreverse intrinsic Follow the naming convention that bswap uses since it's a similar sort of operation. llvm-svn: 259671
2016-02-04 01:49:38 +08:00
P(bitreverse16, (N));
P(bitreverse32, (N));
P(bitreverse64, (N));
Add two test cases for builtins (mostly related to object size builtins). llvm-svn: 55736
2008-09-04 05:17:21 +08:00
// FIXME
// V(clear_cache, (&N, &N+1));
V(trap, ());
Support constant-evaluation of __builtin_nans* as well as the correct constant evaluation of __builtin_nan*. Most of the work to make this work is in LLVM. Fixes <rdar://problem/7696712> and part of PR 5255. llvm-svn: 97383
2010-02-28 21:00:19 +08:00
R(extract_return_addr, (&N));
PR6515: Implement __builtin_signbit and friends. I'm reasonably sure my implementation is correct, but it would be nice if someone could double-check. llvm-svn: 97864
2010-03-06 10:17:52 +08:00
P(signbit, (1.0));
Add two test cases for builtins (mostly related to object size builtins). llvm-svn: 55736
2008-09-04 05:17:21 +08:00
[Clang] Add __builtin_launder Summary: This patch adds `__builtin_launder`, which is required to implement `std::launder`. Additionally GCC provides `__builtin_launder`, so thing brings Clang in-line with GCC. I'm not exactly sure what magic `__builtin_launder` requires, but based on previous discussions this patch applies a `@llvm.invariant.group.barrier`. As noted in previous discussions, this may not be enough to correctly handle vtables. Reviewers: rnk, majnemer, rsmith Reviewed By: rsmith Subscribers: kristina, Romain-Geissler-1A, erichkeane, amharc, jroelofs, cfe-commits, Prazek Differential Revision: https://reviews.llvm.org/D40218 llvm-svn: 349195
2018-12-15 05:11:28 +08:00
R(launder, (&N));
Add two test cases for builtins (mostly related to object size builtins). llvm-svn: 55736
2008-09-04 05:17:21 +08:00
return 0;
}
switch getBuiltinLibFunction to use the new GetOrCreateLLVMFunction functionality, fixing a crash on the attached testcase. Eliminate the BuiltinFunctions cache, as it can contain dangling pointers. This fixes a bunch of valgrind errors on test/CodeGen/builtins.c llvm-svn: 67484
2009-03-23 05:56:56 +08:00
void foo() {
__builtin_strcat(0, 0);
}
CHECK-LABEL-ify some code gen tests to improve diagnostic experience when tests fail. llvm-svn: 188447
2013-08-15 14:47:53 +08:00
// CHECK-LABEL: define void @bar(
Support constant-evaluation of __builtin_nans* as well as the correct constant evaluation of __builtin_nan*. Most of the work to make this work is in LLVM. Fixes <rdar://problem/7696712> and part of PR 5255. llvm-svn: 97383
2010-02-28 21:00:19 +08:00
void bar() {
float f;
double d;
long double ld;
// LLVM's hex representation of float constants is really unfortunate;
// basically it does a float-to-double "conversion" and then prints the
fix a bunch of comment typos found by codespell. Patch by Luis Felipe Strano Moraes! llvm-svn: 129559
2011-04-15 13:22:18 +08:00
// hex form of that. That gives us weird artifacts like exponents
Support constant-evaluation of __builtin_nans* as well as the correct constant evaluation of __builtin_nan*. Most of the work to make this work is in LLVM. Fixes <rdar://problem/7696712> and part of PR 5255. llvm-svn: 97383
2010-02-28 21:00:19 +08:00
// that aren't numerically similar to the original exponent and
// significand bit-patterns that are offset by three bits (because
// the exponent was expanded from 8 bits to 11).
//
// 0xAE98 == 1010111010011000
// 0x15D3 == 1010111010011
f = __builtin_huge_valf(); // CHECK: float 0x7FF0000000000000
d = __builtin_huge_val(); // CHECK: double 0x7FF0000000000000
ld = __builtin_huge_vall(); // CHECK: x86_fp80 0xK7FFF8000000000000000
f = __builtin_nanf(""); // CHECK: float 0x7FF8000000000000
d = __builtin_nan(""); // CHECK: double 0x7FF8000000000000
ld = __builtin_nanl(""); // CHECK: x86_fp80 0xK7FFFC000000000000000
f = __builtin_nanf("0xAE98"); // CHECK: float 0x7FF815D300000000
d = __builtin_nan("0xAE98"); // CHECK: double 0x7FF800000000AE98
ld = __builtin_nanl("0xAE98"); // CHECK: x86_fp80 0xK7FFFC00000000000AE98
f = __builtin_nansf(""); // CHECK: float 0x7FF4000000000000
d = __builtin_nans(""); // CHECK: double 0x7FF4000000000000
ld = __builtin_nansl(""); // CHECK: x86_fp80 0xK7FFFA000000000000000
f = __builtin_nansf("0xAE98"); // CHECK: float 0x7FF015D300000000
d = __builtin_nans("0xAE98"); // CHECK: double 0x7FF000000000AE98
ld = __builtin_nansl("0xAE98");// CHECK: x86_fp80 0xK7FFF800000000000AE98
}
// CHECK: }
optimize builtin_isnan/isinf to not do an extraneous extension from float -> double (which happens because they are modelled as int(...) functions), and add a testcase for isinf. llvm-svn: 103167
2010-05-06 13:50:07 +08:00
Fix a bug with the use of __builtin_bzero in a conditional expression. Patch by Bharathi Seshadri! llvm-svn: 317776
2017-11-09 17:32:32 +08:00
// CHECK-LABEL: define void @test_conditional_bzero
void test_conditional_bzero() {
char dst[20];
int _sz = 20, len = 20;
return (_sz
? ((_sz >= len)
? __builtin_bzero(dst, len)
: foo())
: __builtin_bzero(dst, len));
// CHECK: call void @llvm.memset
// CHECK: call void @llvm.memset
// CHECK-NOT: phi
}
optimize builtin_isnan/isinf to not do an extraneous extension from float -> double (which happens because they are modelled as int(...) functions), and add a testcase for isinf. llvm-svn: 103167
2010-05-06 13:50:07 +08:00
CHECK-LABEL-ify some code gen tests to improve diagnostic experience when tests fail. llvm-svn: 188447
2013-08-15 14:47:53 +08:00
// CHECK-LABEL: define void @test_float_builtins
implement codegen support for __builtin_isfinite, part of PR6083 llvm-svn: 103168
2010-05-06 14:04:13 +08:00
void test_float_builtins(float F, double D, long double LD) {
optimize builtin_isnan/isinf to not do an extraneous extension from float -> double (which happens because they are modelled as int(...) functions), and add a testcase for isinf. llvm-svn: 103167
2010-05-06 13:50:07 +08:00
volatile int res;
res = __builtin_isinf(F);
CGBuiltin: Use @llvm.fabs rather than fabs libcall when emitting builtins Using the intrinsic allows the SelectionDAGBuilder to turn this call into the FABS Node and also the intrinsic is something the vectorizer knows how to vectorize. This patch also sets the readnone attribute on this call, which should enable additional optmizations. llvm-svn: 217042
2014-09-03 23:24:29 +08:00
// CHECK: call float @llvm.fabs.f32(float
optimize builtin_isnan/isinf to not do an extraneous extension from float -> double (which happens because they are modelled as int(...) functions), and add a testcase for isinf. llvm-svn: 103167
2010-05-06 13:50:07 +08:00
// CHECK: fcmp oeq float {{.*}}, 0x7FF0000000000000
res = __builtin_isinf(D);
CGBuiltin: Use @llvm.fabs rather than fabs libcall when emitting builtins Using the intrinsic allows the SelectionDAGBuilder to turn this call into the FABS Node and also the intrinsic is something the vectorizer knows how to vectorize. This patch also sets the readnone attribute on this call, which should enable additional optmizations. llvm-svn: 217042
2014-09-03 23:24:29 +08:00
// CHECK: call double @llvm.fabs.f64(double
optimize builtin_isnan/isinf to not do an extraneous extension from float -> double (which happens because they are modelled as int(...) functions), and add a testcase for isinf. llvm-svn: 103167
2010-05-06 13:50:07 +08:00
// CHECK: fcmp oeq double {{.*}}, 0x7FF0000000000000
[Modules] Implement __builtin_isinf_sign in Clang. Somehow, we never managed to implement this fully. We could constant fold it like crazy, including constant folding complex arguments, etc. But if you actually needed to generate code for it, error. I've implemented it using the somewhat obvious lowering. Happy for suggestions on a more clever way to lower this. Now, what you might ask does this have to do with modules? Fun story. So it turns out that libstdc++ actually uses __builtin_isinf_sign to implement std::isinf when in C++98 mode, but only inside of a template. So if we're lucky, and we never instantiate that, everything is good. But once we try to instantiate that template function, we need this builtin. All of my customers at least are using C++11 and so they never hit this code path. But what does that have to do with modules? Fun story. So it turns out that with modules we actually observe a bunch of bugs in libstdc++ where their <cmath> header clobbers things exposed by <math.h>. To fix these, we have to provide global function definitions to replace the macros that C99 would have used. And it turns out that ::isinf needs to be implemented using the exact semantics used by the C++98 variant of std::isinf. And so I started to fix this bug in libstdc++ and ceased to be able to compile libstdc++ with Clang. The yaks are legion. llvm-svn: 232778
2015-03-20 06:39:51 +08:00
optimize builtin_isnan/isinf to not do an extraneous extension from float -> double (which happens because they are modelled as int(...) functions), and add a testcase for isinf. llvm-svn: 103167
2010-05-06 13:50:07 +08:00
res = __builtin_isinf(LD);
CGBuiltin: Use @llvm.fabs rather than fabs libcall when emitting builtins Using the intrinsic allows the SelectionDAGBuilder to turn this call into the FABS Node and also the intrinsic is something the vectorizer knows how to vectorize. This patch also sets the readnone attribute on this call, which should enable additional optmizations. llvm-svn: 217042
2014-09-03 23:24:29 +08:00
// CHECK: call x86_fp80 @llvm.fabs.f80(x86_fp80
optimize builtin_isnan/isinf to not do an extraneous extension from float -> double (which happens because they are modelled as int(...) functions), and add a testcase for isinf. llvm-svn: 103167
2010-05-06 13:50:07 +08:00
// CHECK: fcmp oeq x86_fp80 {{.*}}, 0xK7FFF8000000000000000
[Modules] Implement __builtin_isinf_sign in Clang. Somehow, we never managed to implement this fully. We could constant fold it like crazy, including constant folding complex arguments, etc. But if you actually needed to generate code for it, error. I've implemented it using the somewhat obvious lowering. Happy for suggestions on a more clever way to lower this. Now, what you might ask does this have to do with modules? Fun story. So it turns out that libstdc++ actually uses __builtin_isinf_sign to implement std::isinf when in C++98 mode, but only inside of a template. So if we're lucky, and we never instantiate that, everything is good. But once we try to instantiate that template function, we need this builtin. All of my customers at least are using C++11 and so they never hit this code path. But what does that have to do with modules? Fun story. So it turns out that with modules we actually observe a bunch of bugs in libstdc++ where their <cmath> header clobbers things exposed by <math.h>. To fix these, we have to provide global function definitions to replace the macros that C99 would have used. And it turns out that ::isinf needs to be implemented using the exact semantics used by the C++98 variant of std::isinf. And so I started to fix this bug in libstdc++ and ceased to be able to compile libstdc++ with Clang. The yaks are legion. llvm-svn: 232778
2015-03-20 06:39:51 +08:00
res = __builtin_isinf_sign(F);
// CHECK: %[[ABS:.*]] = call float @llvm.fabs.f32(float %[[ARG:.*]])
// CHECK: %[[ISINF:.*]] = fcmp oeq float %[[ABS]], 0x7FF0000000000000
// CHECK: %[[BITCAST:.*]] = bitcast float %[[ARG]] to i32
// CHECK: %[[ISNEG:.*]] = icmp slt i32 %[[BITCAST]], 0
// CHECK: %[[SIGN:.*]] = select i1 %[[ISNEG]], i32 -1, i32 1
// CHECK: select i1 %[[ISINF]], i32 %[[SIGN]], i32 0
res = __builtin_isinf_sign(D);
// CHECK: %[[ABS:.*]] = call double @llvm.fabs.f64(double %[[ARG:.*]])
// CHECK: %[[ISINF:.*]] = fcmp oeq double %[[ABS]], 0x7FF0000000000000
// CHECK: %[[BITCAST:.*]] = bitcast double %[[ARG]] to i64
// CHECK: %[[ISNEG:.*]] = icmp slt i64 %[[BITCAST]], 0
// CHECK: %[[SIGN:.*]] = select i1 %[[ISNEG]], i32 -1, i32 1
// CHECK: select i1 %[[ISINF]], i32 %[[SIGN]], i32 0
res = __builtin_isinf_sign(LD);
// CHECK: %[[ABS:.*]] = call x86_fp80 @llvm.fabs.f80(x86_fp80 %[[ARG:.*]])
// CHECK: %[[ISINF:.*]] = fcmp oeq x86_fp80 %[[ABS]], 0xK7FFF8000000000000000
// CHECK: %[[BITCAST:.*]] = bitcast x86_fp80 %[[ARG]] to i80
// CHECK: %[[ISNEG:.*]] = icmp slt i80 %[[BITCAST]], 0
// CHECK: %[[SIGN:.*]] = select i1 %[[ISNEG]], i32 -1, i32 1
// CHECK: select i1 %[[ISINF]], i32 %[[SIGN]], i32 0
implement codegen support for __builtin_isfinite, part of PR6083 llvm-svn: 103168
2010-05-06 14:04:13 +08:00
res = __builtin_isfinite(F);
CGBuiltin: Use @llvm.fabs rather than fabs libcall when emitting builtins Using the intrinsic allows the SelectionDAGBuilder to turn this call into the FABS Node and also the intrinsic is something the vectorizer knows how to vectorize. This patch also sets the readnone attribute on this call, which should enable additional optmizations. llvm-svn: 217042
2014-09-03 23:24:29 +08:00
// CHECK: call float @llvm.fabs.f32(float
make __builtin_isfinite more efficient (PR27145) isinf (is infinite) and isfinite should be implemented with the same function except we change the comparison operator. See PR27145 for more details: https://llvm.org/bugs/show_bug.cgi?id=27145 Ref: forked off of the discussion in D18513. Differential Revision: http://reviews.llvm.org/D18648 llvm-svn: 265675
2016-04-07 22:29:05 +08:00
// CHECK: fcmp one float {{.*}}, 0x7FF0000000000000
Implement codegen for __builtin_isnormal. llvm-svn: 104118
2010-05-19 19:24:26 +08:00
Convert finite to builtin Summary: This patch converts finite/__finite to builtin functions so that it will be inlined by compiler. Reviewers: hfinkel, davidxl, efriedma Subscribers: efriedma, llvm-commits Differential Revision: https://reviews.llvm.org/D24483 llvm-svn: 281509
2016-09-15 01:34:14 +08:00
res = finite(D);
// CHECK: call double @llvm.fabs.f64(double
// CHECK: fcmp one double {{.*}}, 0x7FF0000000000000
Implement codegen for __builtin_isnormal. llvm-svn: 104118
2010-05-19 19:24:26 +08:00
res = __builtin_isnormal(F);
// CHECK: fcmp oeq float
CGBuiltin: Use @llvm.fabs rather than fabs libcall when emitting builtins Using the intrinsic allows the SelectionDAGBuilder to turn this call into the FABS Node and also the intrinsic is something the vectorizer knows how to vectorize. This patch also sets the readnone attribute on this call, which should enable additional optmizations. llvm-svn: 217042
2014-09-03 23:24:29 +08:00
// CHECK: call float @llvm.fabs.f32(float
Implement codegen for __builtin_isnormal. llvm-svn: 104118
2010-05-19 19:24:26 +08:00
// CHECK: fcmp ult float {{.*}}, 0x7FF0000000000000
// CHECK: fcmp uge float {{.*}}, 0x3810000000000000
// CHECK: and i1
// CHECK: and i1
optimize builtin_isnan/isinf to not do an extraneous extension from float -> double (which happens because they are modelled as int(...) functions), and add a testcase for isinf. llvm-svn: 103167
2010-05-06 13:50:07 +08:00
}
CGBuiltin: Use frem instruction rather than libcall to implement fmod AFAICT the semantics of frem match libm's fmod. Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu> Reviewed-by: Tom Stellard <tom@stellard.net> llvm-svn: 218488
2014-09-26 09:19:41 +08:00
// CHECK-LABEL: define void @test_float_builtin_ops
void test_float_builtin_ops(float F, double D, long double LD) {
volatile float resf;
volatile double resd;
volatile long double resld;
resf = __builtin_fmodf(F,F);
// CHECK: frem float
resd = __builtin_fmod(D,D);
// CHECK: frem double
resld = __builtin_fmodl(LD,LD);
// CHECK: frem x86_fp80
Lower __builtin_fabs* to @llvm.fabs.* mingw64's headers implement fabs by calling __builtin_fabs, so using the library call results in an infinite loop. If the backend legalizes @llvm.fabs as a call to fabs later, things should work out, as the crt provides a definition. llvm-svn: 221206
2014-11-04 07:52:09 +08:00
resf = __builtin_fabsf(F);
resd = __builtin_fabs(D);
resld = __builtin_fabsl(LD);
// CHECK: call float @llvm.fabs.f32(float
// CHECK: call double @llvm.fabs.f64(double
// CHECK: call x86_fp80 @llvm.fabs.f80(x86_fp80
Add __builtin_canonicalize llvm-svn: 262122
2016-02-27 17:06:18 +08:00
resf = __builtin_canonicalizef(F);
resd = __builtin_canonicalize(D);
resld = __builtin_canonicalizel(LD);
// CHECK: call float @llvm.canonicalize.f32(float
// CHECK: call double @llvm.canonicalize.f64(double
// CHECK: call x86_fp80 @llvm.canonicalize.f80(x86_fp80
Emit more intrinsics for builtin functions This is important for building libclc. Since r273039 tests are failing due to now emitting calls to these functions instead of emitting the DAG node. The libm function names are implemented for OpenCL, and should call the locally defined versions, so -fno-builtin is used. The IR Some functions use the __builtins and expect the intrinsics to be emitted. Without this we end up with nobuiltin calls to intrinsics or to unsupported library calls. llvm-svn: 274370
2016-07-02 01:38:14 +08:00
resf = __builtin_fminf(F, F);
// CHECK: call float @llvm.minnum.f32
resd = __builtin_fmin(D, D);
// CHECK: call double @llvm.minnum.f64
resld = __builtin_fminl(LD, LD);
// CHECK: call x86_fp80 @llvm.minnum.f80
resf = __builtin_fmaxf(F, F);
// CHECK: call float @llvm.maxnum.f32
resd = __builtin_fmax(D, D);
// CHECK: call double @llvm.maxnum.f64
resld = __builtin_fmaxl(LD, LD);
// CHECK: call x86_fp80 @llvm.maxnum.f80
resf = __builtin_fabsf(F);
// CHECK: call float @llvm.fabs.f32
resd = __builtin_fabs(D);
// CHECK: call double @llvm.fabs.f64
resld = __builtin_fabsl(LD);
// CHECK: call x86_fp80 @llvm.fabs.f80
resf = __builtin_copysignf(F, F);
// CHECK: call float @llvm.copysign.f32
resd = __builtin_copysign(D, D);
// CHECK: call double @llvm.copysign.f64
resld = __builtin_copysignl(LD, LD);
// CHECK: call x86_fp80 @llvm.copysign.f80
resf = __builtin_ceilf(F);
// CHECK: call float @llvm.ceil.f32
resd = __builtin_ceil(D);
// CHECK: call double @llvm.ceil.f64
resld = __builtin_ceill(LD);
// CHECK: call x86_fp80 @llvm.ceil.f80
resf = __builtin_floorf(F);
// CHECK: call float @llvm.floor.f32
resd = __builtin_floor(D);
// CHECK: call double @llvm.floor.f64
resld = __builtin_floorl(LD);
// CHECK: call x86_fp80 @llvm.floor.f80
[CodeGen] add tests for __builtin_sqrt*; NFC I don't know if this is correct, but this is what we currently do. More discussion in PR27108 and PR27435 and D27618. llvm-svn: 316250
2017-10-21 07:32:41 +08:00
resf = __builtin_sqrtf(F);
[CodeGen] convert math libcalls/builtins to equivalent LLVM intrinsics 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
2017-12-02 07:15:52 +08:00
// CHECK: call float @llvm.sqrt.f32(
[CodeGen] add tests for __builtin_sqrt*; NFC I don't know if this is correct, but this is what we currently do. More discussion in PR27108 and PR27435 and D27618. llvm-svn: 316250
2017-10-21 07:32:41 +08:00
resd = __builtin_sqrt(D);
[CodeGen] convert math libcalls/builtins to equivalent LLVM intrinsics 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
2017-12-02 07:15:52 +08:00
// CHECK: call double @llvm.sqrt.f64(
[CodeGen] add tests for __builtin_sqrt*; NFC I don't know if this is correct, but this is what we currently do. More discussion in PR27108 and PR27435 and D27618. llvm-svn: 316250
2017-10-21 07:32:41 +08:00
resld = __builtin_sqrtl(LD);
[CodeGen] convert math libcalls/builtins to equivalent LLVM intrinsics 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
2017-12-02 07:15:52 +08:00
// CHECK: call x86_fp80 @llvm.sqrt.f80
[CodeGen] add tests for __builtin_sqrt*; NFC I don't know if this is correct, but this is what we currently do. More discussion in PR27108 and PR27435 and D27618. llvm-svn: 316250
2017-10-21 07:32:41 +08:00
Emit more intrinsics for builtin functions This is important for building libclc. Since r273039 tests are failing due to now emitting calls to these functions instead of emitting the DAG node. The libm function names are implemented for OpenCL, and should call the locally defined versions, so -fno-builtin is used. The IR Some functions use the __builtins and expect the intrinsics to be emitted. Without this we end up with nobuiltin calls to intrinsics or to unsupported library calls. llvm-svn: 274370
2016-07-02 01:38:14 +08:00
resf = __builtin_truncf(F);
// CHECK: call float @llvm.trunc.f32
resd = __builtin_trunc(D);
// CHECK: call double @llvm.trunc.f64
resld = __builtin_truncl(LD);
// CHECK: call x86_fp80 @llvm.trunc.f80
resf = __builtin_rintf(F);
// CHECK: call float @llvm.rint.f32
resd = __builtin_rint(D);
// CHECK: call double @llvm.rint.f64
resld = __builtin_rintl(LD);
// CHECK: call x86_fp80 @llvm.rint.f80
resf = __builtin_nearbyintf(F);
// CHECK: call float @llvm.nearbyint.f32
resd = __builtin_nearbyint(D);
// CHECK: call double @llvm.nearbyint.f64
resld = __builtin_nearbyintl(LD);
// CHECK: call x86_fp80 @llvm.nearbyint.f80
resf = __builtin_roundf(F);
// CHECK: call float @llvm.round.f32
resd = __builtin_round(D);
// CHECK: call double @llvm.round.f64
resld = __builtin_roundl(LD);
// CHECK: call x86_fp80 @llvm.round.f80
CGBuiltin: Use frem instruction rather than libcall to implement fmod AFAICT the semantics of frem match libm's fmod. Signed-off-by: Jan Vesely <jan.vesely@rutgers.edu> Reviewed-by: Tom Stellard <tom@stellard.net> llvm-svn: 218488
2014-09-26 09:19:41 +08:00
}
test/CodeGen/builtins.c: reinstate #ifdef __x86_64__ around __builtin_longjmp Unadvertently removed in r285019 llvm-svn: 285041
2016-10-25 07:38:24 +08:00
// __builtin_longjmp isn't supported on all platforms, so only test it on X86.
#ifdef __x86_64__
CHECK-LABEL-ify some code gen tests to improve diagnostic experience when tests fail. llvm-svn: 188447
2013-08-15 14:47:53 +08:00
// CHECK-LABEL: define void @test_builtin_longjmp
Enable the implementation of __builtin_setjmp and __builtin_longjmp. Not all LLVM backends support these yet. llvm-svn: 104867
2010-05-28 02:47:06 +08:00
void test_builtin_longjmp(void **buffer) {
// CHECK: [[BITCAST:%.*]] = bitcast
// CHECK-NEXT: call void @llvm.eh.sjlj.longjmp(i8* [[BITCAST]])
__builtin_longjmp(buffer, 1);
// CHECK-NEXT: unreachable
}
Add __builtin_readcyclecounter() to produce the @llvm.readcyclecounter() intrinsic. llvm-svn: 161310
2012-08-06 06:03:08 +08:00
test/CodeGen/builtins.c: reinstate #ifdef __x86_64__ around __builtin_longjmp Unadvertently removed in r285019 llvm-svn: 285041
2016-10-25 07:38:24 +08:00
#endif
CHECK-LABEL-ify some code gen tests to improve diagnostic experience when tests fail. llvm-svn: 188447
2013-08-15 14:47:53 +08:00
// CHECK-LABEL: define i64 @test_builtin_readcyclecounter
Add __builtin_readcyclecounter() to produce the @llvm.readcyclecounter() intrinsic. llvm-svn: 161310
2012-08-06 06:03:08 +08:00
long long test_builtin_readcyclecounter() {
// CHECK: call i64 @llvm.readcyclecounter()
return __builtin_readcyclecounter();
}
Add support for __builtin_os_log_format[_buffer_size] This reverts commit r285007 and reapply r284990, with a fix for the opencl test that I broke. Original commit message follows: These new builtins support a mechanism for logging OS events, using a printf-like format string to specify the layout of data in a buffer. The _buffer_size version of the builtin can be used to determine the size of the buffer to allocate to hold the data, and then __builtin_os_log_format can write data into that buffer. This implements format checking to report mismatches between the format string and the data arguments. Most of this code was written by Chris Willmore. Differential Revision: https://reviews.llvm.org/D25888 llvm-svn: 285019
2016-10-25 04:39:34 +08:00
[Clang] Add __builtin_launder Summary: This patch adds `__builtin_launder`, which is required to implement `std::launder`. Additionally GCC provides `__builtin_launder`, so thing brings Clang in-line with GCC. I'm not exactly sure what magic `__builtin_launder` requires, but based on previous discussions this patch applies a `@llvm.invariant.group.barrier`. As noted in previous discussions, this may not be enough to correctly handle vtables. Reviewers: rnk, majnemer, rsmith Reviewed By: rsmith Subscribers: kristina, Romain-Geissler-1A, erichkeane, amharc, jroelofs, cfe-commits, Prazek Differential Revision: https://reviews.llvm.org/D40218 llvm-svn: 349195
2018-12-15 05:11:28 +08:00
/// __builtin_launder should be a NOP in C since there are no vtables.
// CHECK-LABEL: define void @test_builtin_launder
void test_builtin_launder(int *p) {
// CHECK: [[TMP:%.*]] = load i32*,
// CHECK-NOT: @llvm.launder
// CHECK: store i32* [[TMP]],
int *d = __builtin_launder(p);
}
Fix test on non-X86 platforms This is a fixup for r285019, adding an `#ifdef __x86_64__` since the os_log builtin is platform specific. llvm-svn: 285027
2016-10-25 05:22:01 +08:00
// Behavior of __builtin_os_log differs between platforms, so only test on X86
#ifdef __x86_64__
Add support for __builtin_os_log_format[_buffer_size] This reverts commit r285007 and reapply r284990, with a fix for the opencl test that I broke. Original commit message follows: These new builtins support a mechanism for logging OS events, using a printf-like format string to specify the layout of data in a buffer. The _buffer_size version of the builtin can be used to determine the size of the buffer to allocate to hold the data, and then __builtin_os_log_format can write data into that buffer. This implements format checking to report mismatches between the format string and the data arguments. Most of this code was written by Chris Willmore. Differential Revision: https://reviews.llvm.org/D25888 llvm-svn: 285019
2016-10-25 04:39:34 +08:00
// CHECK-LABEL: define void @test_builtin_os_log
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK: (i8* %[[BUF:.*]], i32 %[[I:.*]], i8* %[[DATA:.*]])
Add support for __builtin_os_log_format[_buffer_size] This reverts commit r285007 and reapply r284990, with a fix for the opencl test that I broke. Original commit message follows: These new builtins support a mechanism for logging OS events, using a printf-like format string to specify the layout of data in a buffer. The _buffer_size version of the builtin can be used to determine the size of the buffer to allocate to hold the data, and then __builtin_os_log_format can write data into that buffer. This implements format checking to report mismatches between the format string and the data arguments. Most of this code was written by Chris Willmore. Differential Revision: https://reviews.llvm.org/D25888 llvm-svn: 285019
2016-10-25 04:39:34 +08:00
void test_builtin_os_log(void *buf, int i, const char *data) {
volatile int len;
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK: %[[BUF_ADDR:.*]] = alloca i8*, align 8
// CHECK: %[[I_ADDR:.*]] = alloca i32, align 4
// CHECK: %[[DATA_ADDR:.*]] = alloca i8*, align 8
// CHECK: %[[LEN:.*]] = alloca i32, align 4
// CHECK: store i8* %[[BUF]], i8** %[[BUF_ADDR]], align 8
// CHECK: store i32 %[[I]], i32* %[[I_ADDR]], align 4
// CHECK: store i8* %[[DATA]], i8** %[[DATA_ADDR]], align 8
// CHECK: store volatile i32 34, i32* %[[LEN]]
Add support for __builtin_os_log_format[_buffer_size] This reverts commit r285007 and reapply r284990, with a fix for the opencl test that I broke. Original commit message follows: These new builtins support a mechanism for logging OS events, using a printf-like format string to specify the layout of data in a buffer. The _buffer_size version of the builtin can be used to determine the size of the buffer to allocate to hold the data, and then __builtin_os_log_format can write data into that buffer. This implements format checking to report mismatches between the format string and the data arguments. Most of this code was written by Chris Willmore. Differential Revision: https://reviews.llvm.org/D25888 llvm-svn: 285019
2016-10-25 04:39:34 +08:00
len = __builtin_os_log_format_buffer_size("%d %{public}s %{private}.16P", i, data, data);
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK: %[[V1:.*]] = load i8*, i8** %[[BUF_ADDR]]
// CHECK: %[[V2:.*]] = load i32, i32* %[[I_ADDR]]
// CHECK: %[[V3:.*]] = load i8*, i8** %[[DATA_ADDR]]
// CHECK: %[[V4:.*]] = ptrtoint i8* %[[V3]] to i64
// CHECK: %[[V5:.*]] = load i8*, i8** %[[DATA_ADDR]]
// CHECK: %[[V6:.*]] = ptrtoint i8* %[[V5]] to i64
// CHECK: call void @__os_log_helper_1_3_4_4_0_8_34_4_17_8_49(i8* %[[V1]], i32 %[[V2]], i64 %[[V4]], i32 16, i64 %[[V6]])
os_log: When there are multiple privacy annotations in the format string, choose the strictest one instead of the last. Also fix an undefined behavior. Move the pointer update to a later point to avoid adding StringRef::npos to the pointer. rdar://problem/40706280 llvm-svn: 336863
2018-07-12 06:19:14 +08:00
__builtin_os_log_format(buf, "%d %{public}s %{private}.16P", i, data, data);
Fix parsing of privacy annotations in os_log format strings. Privacy annotations shouldn't have to appear in the first comma-delimited string in order to be recognized. Also, they should be ignored if they are preceded or followed by non-whitespace characters. rdar://problem/40706280 llvm-svn: 336629
2018-07-10 08:50:25 +08:00
// privacy annotations aren't recognized when they are preceded or followed
// by non-whitespace characters.
// CHECK: call void @__os_log_helper_1_2_1_8_32(
__builtin_os_log_format(buf, "%{xyz public}s", data);
// CHECK: call void @__os_log_helper_1_2_1_8_32(
__builtin_os_log_format(buf, "%{ public xyz}s", data);
// CHECK: call void @__os_log_helper_1_2_1_8_32(
__builtin_os_log_format(buf, "%{ public1}s", data);
// Privacy annotations do not have to be in the first comma-delimited string.
// CHECK: call void @__os_log_helper_1_2_1_8_34(
__builtin_os_log_format(buf, "%{ xyz, public }s", "abc");
os_log: Minor code cleanups. NFC. Also, add a new test case and fix an incorrect comment. llvm-svn: 346209
2018-11-06 13:41:33 +08:00
// CHECK: call void @__os_log_helper_1_3_1_8_33(
__builtin_os_log_format(buf, "%{ xyz, private }s", "abc");
os_log: Add a new privacy annotation "sensitive". This is a stricter privacy annotation than "private", which will be used for data that shouldn’t be logged to disk. For backward compatibility, the "private" bit is set too. rdar://problem/36755912 llvm-svn: 346210
2018-11-06 14:26:17 +08:00
// CHECK: call void @__os_log_helper_1_3_1_8_37(
__builtin_os_log_format(buf, "%{ xyz, sensitive }s", "abc");
os_log: Minor code cleanups. NFC. Also, add a new test case and fix an incorrect comment. llvm-svn: 346209
2018-11-06 13:41:33 +08:00
// The strictest privacy annotation in the string wins.
Fix parsing of privacy annotations in os_log format strings. Privacy annotations shouldn't have to appear in the first comma-delimited string in order to be recognized. Also, they should be ignored if they are preceded or followed by non-whitespace characters. rdar://problem/40706280 llvm-svn: 336629
2018-07-10 08:50:25 +08:00
// CHECK: call void @__os_log_helper_1_3_1_8_33(
os_log: When there are multiple privacy annotations in the format string, choose the strictest one instead of the last. Also fix an undefined behavior. Move the pointer update to a later point to avoid adding StringRef::npos to the pointer. rdar://problem/40706280 llvm-svn: 336863
2018-07-12 06:19:14 +08:00
__builtin_os_log_format(buf, "%{ private, public, private, public}s", "abc");
os_log: Add a new privacy annotation "sensitive". This is a stricter privacy annotation than "private", which will be used for data that shouldn’t be logged to disk. For backward compatibility, the "private" bit is set too. rdar://problem/36755912 llvm-svn: 346210
2018-11-06 14:26:17 +08:00
// CHECK: call void @__os_log_helper_1_3_1_8_37(
__builtin_os_log_format(buf, "%{ private, sensitive, private, public}s",
"abc");
os_log: Allow specifying mask type in format string. A mask type is a 1 to 8-byte string that follows the "mask." annotation in the format string. This enables obfuscating data in the event the provided privacy level isn't enabled. rdar://problem/36756282 llvm-svn: 346211
2018-11-06 15:05:14 +08:00
// CHECK: store volatile i32 22, i32* %[[LEN]], align 4
len = __builtin_os_log_format_buffer_size("%{mask.xyz}s", "abc");
// CHECK: call void @__os_log_helper_1_2_2_8_112_8_34(i8* {{.*}}, i64 8026488
__builtin_os_log_format(buf, "%{mask.xyz, public}s", "abc");
// CHECK: call void @__os_log_helper_1_3_2_8_112_4_1(i8* {{.*}}, i64 8026488
__builtin_os_log_format(buf, "%{ mask.xyz, private }d", 11);
// Mask type is silently ignored.
// CHECK: call void @__os_log_helper_1_2_1_8_32(
__builtin_os_log_format(buf, "%{ mask. xyz }s", "abc");
// CHECK: call void @__os_log_helper_1_2_1_8_32(
__builtin_os_log_format(buf, "%{ mask.xy z }s", "abc");
Add support for __builtin_os_log_format[_buffer_size] This reverts commit r285007 and reapply r284990, with a fix for the opencl test that I broke. Original commit message follows: These new builtins support a mechanism for logging OS events, using a printf-like format string to specify the layout of data in a buffer. The _buffer_size version of the builtin can be used to determine the size of the buffer to allocate to hold the data, and then __builtin_os_log_format can write data into that buffer. This implements format checking to report mismatches between the format string and the data arguments. Most of this code was written by Chris Willmore. Differential Revision: https://reviews.llvm.org/D25888 llvm-svn: 285019
2016-10-25 04:39:34 +08:00
}
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK-LABEL: define linkonce_odr hidden void @__os_log_helper_1_3_4_4_0_8_34_4_17_8_49
// CHECK: (i8* %[[BUFFER:.*]], i32 %[[ARG0:.*]], i64 %[[ARG1:.*]], i32 %[[ARG2:.*]], i64 %[[ARG3:.*]])
// CHECK: %[[BUFFER_ADDR:.*]] = alloca i8*, align 8
// CHECK: %[[ARG0_ADDR:.*]] = alloca i32, align 4
// CHECK: %[[ARG1_ADDR:.*]] = alloca i64, align 8
// CHECK: %[[ARG2_ADDR:.*]] = alloca i32, align 4
// CHECK: %[[ARG3_ADDR:.*]] = alloca i64, align 8
// CHECK: store i8* %[[BUFFER]], i8** %[[BUFFER_ADDR]], align 8
// CHECK: store i32 %[[ARG0]], i32* %[[ARG0_ADDR]], align 4
// CHECK: store i64 %[[ARG1]], i64* %[[ARG1_ADDR]], align 8
// CHECK: store i32 %[[ARG2]], i32* %[[ARG2_ADDR]], align 4
// CHECK: store i64 %[[ARG3]], i64* %[[ARG3_ADDR]], align 8
// CHECK: %[[BUF:.*]] = load i8*, i8** %[[BUFFER_ADDR]], align 8
// CHECK: %[[SUMMARY:.*]] = getelementptr i8, i8* %[[BUF]], i64 0
// CHECK: store i8 3, i8* %[[SUMMARY]], align 1
// CHECK: %[[NUMARGS:.*]] = getelementptr i8, i8* %[[BUF]], i64 1
// CHECK: store i8 4, i8* %[[NUMARGS]], align 1
// CHECK: %[[ARGDESCRIPTOR:.*]] = getelementptr i8, i8* %[[BUF]], i64 2
// CHECK: store i8 0, i8* %[[ARGDESCRIPTOR]], align 1
// CHECK: %[[ARGSIZE:.*]] = getelementptr i8, i8* %[[BUF]], i64 3
// CHECK: store i8 4, i8* %[[ARGSIZE]], align 1
// CHECK: %[[ARGDATA:.*]] = getelementptr i8, i8* %[[BUF]], i64 4
// CHECK: %[[ARGDATACAST:.*]] = bitcast i8* %[[ARGDATA]] to i32*
// CHECK: %[[V0:.*]] = load i32, i32* %[[ARG0_ADDR]], align 4
// CHECK: store i32 %[[V0]], i32* %[[ARGDATACAST]], align 1
// CHECK: %[[ARGDESCRIPTOR1:.*]] = getelementptr i8, i8* %[[BUF]], i64 8
// CHECK: store i8 34, i8* %[[ARGDESCRIPTOR1]], align 1
// CHECK: %[[ARGSIZE2:.*]] = getelementptr i8, i8* %[[BUF]], i64 9
// CHECK: store i8 8, i8* %[[ARGSIZE2]], align 1
// CHECK: %[[ARGDATA3:.*]] = getelementptr i8, i8* %[[BUF]], i64 10
// CHECK: %[[ARGDATACAST4:.*]] = bitcast i8* %[[ARGDATA3]] to i64*
// CHECK: %[[V1:.*]] = load i64, i64* %[[ARG1_ADDR]], align 8
// CHECK: store i64 %[[V1]], i64* %[[ARGDATACAST4]], align 1
// CHECK: %[[ARGDESCRIPTOR5:.*]] = getelementptr i8, i8* %[[BUF]], i64 18
// CHECK: store i8 17, i8* %[[ARGDESCRIPTOR5]], align 1
// CHECK: %[[ARGSIZE6:.*]] = getelementptr i8, i8* %[[BUF]], i64 19
// CHECK: store i8 4, i8* %[[ARGSIZE6]], align 1
// CHECK: %[[ARGDATA7:.*]] = getelementptr i8, i8* %[[BUF]], i64 20
// CHECK: %[[ARGDATACAST8:.*]] = bitcast i8* %[[ARGDATA7]] to i32*
// CHECK: %[[V2:.*]] = load i32, i32* %[[ARG2_ADDR]], align 4
// CHECK: store i32 %[[V2]], i32* %[[ARGDATACAST8]], align 1
// CHECK: %[[ARGDESCRIPTOR9:.*]] = getelementptr i8, i8* %[[BUF]], i64 24
// CHECK: store i8 49, i8* %[[ARGDESCRIPTOR9]], align 1
// CHECK: %[[ARGSIZE10:.*]] = getelementptr i8, i8* %[[BUF]], i64 25
// CHECK: store i8 8, i8* %[[ARGSIZE10]], align 1
// CHECK: %[[ARGDATA11:.*]] = getelementptr i8, i8* %[[BUF]], i64 26
// CHECK: %[[ARGDATACAST12:.*]] = bitcast i8* %[[ARGDATA11]] to i64*
// CHECK: %[[V3:.*]] = load i64, i64* %[[ARG3_ADDR]], align 8
// CHECK: store i64 %[[V3]], i64* %[[ARGDATACAST12]], align 1
Add support for __builtin_os_log_format[_buffer_size] This reverts commit r285007 and reapply r284990, with a fix for the opencl test that I broke. Original commit message follows: These new builtins support a mechanism for logging OS events, using a printf-like format string to specify the layout of data in a buffer. The _buffer_size version of the builtin can be used to determine the size of the buffer to allocate to hold the data, and then __builtin_os_log_format can write data into that buffer. This implements format checking to report mismatches between the format string and the data arguments. Most of this code was written by Chris Willmore. Differential Revision: https://reviews.llvm.org/D25888 llvm-svn: 285019
2016-10-25 04:39:34 +08:00
// CHECK-LABEL: define void @test_builtin_os_log_wide
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK: (i8* %[[BUF:.*]], i8* %[[DATA:.*]], i32* %[[STR:.*]])
Add support for __builtin_os_log_format[_buffer_size] This reverts commit r285007 and reapply r284990, with a fix for the opencl test that I broke. Original commit message follows: These new builtins support a mechanism for logging OS events, using a printf-like format string to specify the layout of data in a buffer. The _buffer_size version of the builtin can be used to determine the size of the buffer to allocate to hold the data, and then __builtin_os_log_format can write data into that buffer. This implements format checking to report mismatches between the format string and the data arguments. Most of this code was written by Chris Willmore. Differential Revision: https://reviews.llvm.org/D25888 llvm-svn: 285019
2016-10-25 04:39:34 +08:00
typedef int wchar_t;
void test_builtin_os_log_wide(void *buf, const char *data, wchar_t *str) {
volatile int len;
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK: %[[BUF_ADDR:.*]] = alloca i8*, align 8
// CHECK: %[[DATA_ADDR:.*]] = alloca i8*, align 8
// CHECK: %[[STR_ADDR:.*]] = alloca i32*, align 8
// CHECK: %[[LEN:.*]] = alloca i32, align 4
// CHECK: store i8* %[[BUF]], i8** %[[BUF_ADDR]], align 8
// CHECK: store i8* %[[DATA]], i8** %[[DATA_ADDR]], align 8
// CHECK: store i32* %[[STR]], i32** %[[STR_ADDR]], align 8
// CHECK: store volatile i32 12, i32* %[[LEN]], align 4
Add support for __builtin_os_log_format[_buffer_size] This reverts commit r285007 and reapply r284990, with a fix for the opencl test that I broke. Original commit message follows: These new builtins support a mechanism for logging OS events, using a printf-like format string to specify the layout of data in a buffer. The _buffer_size version of the builtin can be used to determine the size of the buffer to allocate to hold the data, and then __builtin_os_log_format can write data into that buffer. This implements format checking to report mismatches between the format string and the data arguments. Most of this code was written by Chris Willmore. Differential Revision: https://reviews.llvm.org/D25888 llvm-svn: 285019
2016-10-25 04:39:34 +08:00
len = __builtin_os_log_format_buffer_size("%S", str);
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK: %[[V1:.*]] = load i8*, i8** %[[BUF_ADDR]], align 8
// CHECK: %[[V2:.*]] = load i32*, i32** %[[STR_ADDR]], align 8
// CHECK: %[[V3:.*]] = ptrtoint i32* %[[V2]] to i64
// CHECK: call void @__os_log_helper_1_2_1_8_80(i8* %[[V1]], i64 %[[V3]])
Add support for __builtin_os_log_format[_buffer_size] This reverts commit r285007 and reapply r284990, with a fix for the opencl test that I broke. Original commit message follows: These new builtins support a mechanism for logging OS events, using a printf-like format string to specify the layout of data in a buffer. The _buffer_size version of the builtin can be used to determine the size of the buffer to allocate to hold the data, and then __builtin_os_log_format can write data into that buffer. This implements format checking to report mismatches between the format string and the data arguments. Most of this code was written by Chris Willmore. Differential Revision: https://reviews.llvm.org/D25888 llvm-svn: 285019
2016-10-25 04:39:34 +08:00
__builtin_os_log_format(buf, "%S", str);
}
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK-LABEL: define linkonce_odr hidden void @__os_log_helper_1_2_1_8_80
// CHECK: (i8* %[[BUFFER:.*]], i64 %[[ARG0:.*]])
// CHECK: %[[BUFFER_ADDR:.*]] = alloca i8*, align 8
// CHECK: %[[ARG0_ADDR:.*]] = alloca i64, align 8
// CHECK: store i8* %[[BUFFER]], i8** %[[BUFFER_ADDR]], align 8
// CHECK: store i64 %[[ARG0]], i64* %[[ARG0_ADDR]], align 8
// CHECK: %[[BUF:.*]] = load i8*, i8** %[[BUFFER_ADDR]], align 8
// CHECK: %[[SUMMARY:.*]] = getelementptr i8, i8* %[[BUF]], i64 0
// CHECK: store i8 2, i8* %[[SUMMARY]], align 1
// CHECK: %[[NUMARGS:.*]] = getelementptr i8, i8* %[[BUF]], i64 1
// CHECK: store i8 1, i8* %[[NUMARGS]], align 1
// CHECK: %[[ARGDESCRIPTOR:.*]] = getelementptr i8, i8* %[[BUF]], i64 2
// CHECK: store i8 80, i8* %[[ARGDESCRIPTOR]], align 1
// CHECK: %[[ARGSIZE:.*]] = getelementptr i8, i8* %[[BUF]], i64 3
// CHECK: store i8 8, i8* %[[ARGSIZE]], align 1
// CHECK: %[[ARGDATA:.*]] = getelementptr i8, i8* %[[BUF]], i64 4
// CHECK: %[[ARGDATACAST:.*]] = bitcast i8* %[[ARGDATA]] to i64*
// CHECK: %[[V0:.*]] = load i64, i64* %[[ARG0_ADDR]], align 8
// CHECK: store i64 %[[V0]], i64* %[[ARGDATACAST]], align 1
Fix os_log formating with arbitrary precision and field width llvm-svn: 289761
2016-12-15 12:02:31 +08:00
// CHECK-LABEL: define void @test_builtin_os_log_precision_width
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK: (i8* %[[BUF:.*]], i8* %[[DATA:.*]], i32 %[[PRECISION:.*]], i32 %[[WIDTH:.*]])
Fix os_log formating with arbitrary precision and field width llvm-svn: 289761
2016-12-15 12:02:31 +08:00
void test_builtin_os_log_precision_width(void *buf, const char *data,
int precision, int width) {
volatile int len;
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK: %[[BUF_ADDR:.*]] = alloca i8*, align 8
// CHECK: %[[DATA_ADDR:.*]] = alloca i8*, align 8
// CHECK: %[[PRECISION_ADDR:.*]] = alloca i32, align 4
// CHECK: %[[WIDTH_ADDR:.*]] = alloca i32, align 4
// CHECK: %[[LEN:.*]] = alloca i32, align 4
// CHECK: store i8* %[[BUF]], i8** %[[BUF_ADDR]], align 8
// CHECK: store i8* %[[DATA]], i8** %[[DATA_ADDR]], align 8
// CHECK: store i32 %[[PRECISION]], i32* %[[PRECISION_ADDR]], align 4
// CHECK: store i32 %[[WIDTH]], i32* %[[WIDTH_ADDR]], align 4
// CHECK: store volatile i32 24, i32* %[[LEN]], align 4
Fix os_log formating with arbitrary precision and field width llvm-svn: 289761
2016-12-15 12:02:31 +08:00
len = __builtin_os_log_format_buffer_size("Hello %*.*s World", precision, width, data);
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK: %[[V1:.*]] = load i8*, i8** %[[BUF_ADDR]], align 8
// CHECK: %[[V2:.*]] = load i32, i32* %[[PRECISION_ADDR]], align 4
// CHECK: %[[V3:.*]] = load i32, i32* %[[WIDTH_ADDR]], align 4
// CHECK: %[[V4:.*]] = load i8*, i8** %[[DATA_ADDR]], align 8
// CHECK: %[[V5:.*]] = ptrtoint i8* %[[V4]] to i64
// CHECK: call void @__os_log_helper_1_2_3_4_0_4_16_8_32(i8* %[[V1]], i32 %[[V2]], i32 %[[V3]], i64 %[[V5]])
Fix os_log formating with arbitrary precision and field width llvm-svn: 289761
2016-12-15 12:02:31 +08:00
__builtin_os_log_format(buf, "Hello %*.*s World", precision, width, data);
}
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK-LABEL: define linkonce_odr hidden void @__os_log_helper_1_2_3_4_0_4_16_8_32
// CHECK: (i8* %[[BUFFER:.*]], i32 %[[ARG0:.*]], i32 %[[ARG1:.*]], i64 %[[ARG2:.*]])
// CHECK: %[[BUFFER_ADDR:.*]] = alloca i8*, align 8
// CHECK: %[[ARG0_ADDR:.*]] = alloca i32, align 4
// CHECK: %[[ARG1_ADDR:.*]] = alloca i32, align 4
// CHECK: %[[ARG2_ADDR:.*]] = alloca i64, align 8
// CHECK: store i8* %[[BUFFER]], i8** %[[BUFFER_ADDR]], align 8
// CHECK: store i32 %[[ARG0]], i32* %[[ARG0_ADDR]], align 4
// CHECK: store i32 %[[ARG1]], i32* %[[ARG1_ADDR]], align 4
// CHECK: store i64 %[[ARG2]], i64* %[[ARG2_ADDR]], align 8
// CHECK: %[[BUF:.*]] = load i8*, i8** %[[BUFFER_ADDR]], align 8
// CHECK: %[[SUMMARY:.*]] = getelementptr i8, i8* %[[BUF]], i64 0
// CHECK: store i8 2, i8* %[[SUMMARY]], align 1
// CHECK: %[[NUMARGS:.*]] = getelementptr i8, i8* %[[BUF]], i64 1
// CHECK: store i8 3, i8* %[[NUMARGS]], align 1
// CHECK: %[[ARGDESCRIPTOR:.*]] = getelementptr i8, i8* %[[BUF]], i64 2
// CHECK: store i8 0, i8* %[[ARGDESCRIPTOR]], align 1
// CHECK: %[[ARGSIZE:.*]] = getelementptr i8, i8* %[[BUF]], i64 3
// CHECK: store i8 4, i8* %[[ARGSIZE]], align 1
// CHECK: %[[ARGDATA:.*]] = getelementptr i8, i8* %[[BUF]], i64 4
// CHECK: %[[ARGDATACAST:.*]] = bitcast i8* %[[ARGDATA]] to i32*
// CHECK: %[[V0:.*]] = load i32, i32* %[[ARG0_ADDR]], align 4
// CHECK: store i32 %[[V0]], i32* %[[ARGDATACAST]], align 1
// CHECK: %[[ARGDESCRIPTOR1:.*]] = getelementptr i8, i8* %[[BUF]], i64 8
// CHECK: store i8 16, i8* %[[ARGDESCRIPTOR1]], align 1
// CHECK: %[[ARGSIZE2:.*]] = getelementptr i8, i8* %[[BUF]], i64 9
// CHECK: store i8 4, i8* %[[ARGSIZE2]], align 1
// CHECK: %[[ARGDATA3:.*]] = getelementptr i8, i8* %[[BUF]], i64 10
// CHECK: %[[ARGDATACAST4:.*]] = bitcast i8* %[[ARGDATA3]] to i32*
// CHECK: %[[V1:.*]] = load i32, i32* %[[ARG1_ADDR]], align 4
// CHECK: store i32 %[[V1]], i32* %[[ARGDATACAST4]], align 1
// CHECK: %[[ARGDESCRIPTOR5:.*]] = getelementptr i8, i8* %[[BUF]], i64 14
// CHECK: store i8 32, i8* %[[ARGDESCRIPTOR5]], align 1
// CHECK: %[[ARGSIZE6:.*]] = getelementptr i8, i8* %[[BUF]], i64 15
// CHECK: store i8 8, i8* %[[ARGSIZE6]], align 1
// CHECK: %[[ARGDATA7:.*]] = getelementptr i8, i8* %[[BUF]], i64 16
// CHECK: %[[ARGDATACAST8:.*]] = bitcast i8* %[[ARGDATA7]] to i64*
// CHECK: %[[V2:.*]] = load i64, i64* %[[ARG2_ADDR]], align 8
// CHECK: store i64 %[[V2]], i64* %[[ARGDATACAST8]], align 1
Fix printf specifier handling: invalid specifier should not be marked as "consuming data arguments" Reviewers: rsmith, bruno, dexonsmith Subscribers: cfe-commits Differential Revision: https://reviews.llvm.org/D27796 llvm-svn: 289850
2016-12-16 02:54:00 +08:00
// CHECK-LABEL: define void @test_builtin_os_log_invalid
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK: (i8* %[[BUF:.*]], i32 %[[DATA:.*]])
Fix printf specifier handling: invalid specifier should not be marked as "consuming data arguments" Reviewers: rsmith, bruno, dexonsmith Subscribers: cfe-commits Differential Revision: https://reviews.llvm.org/D27796 llvm-svn: 289850
2016-12-16 02:54:00 +08:00
void test_builtin_os_log_invalid(void *buf, int data) {
volatile int len;
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK: %[[BUF_ADDR:.*]] = alloca i8*, align 8
// CHECK: %[[DATA_ADDR:.*]] = alloca i32, align 4
// CHECK: %[[LEN:.*]] = alloca i32, align 4
// CHECK: store i8* %[[BUF]], i8** %[[BUF_ADDR]], align 8
// CHECK: store i32 %[[DATA]], i32* %[[DATA_ADDR]], align 4
Fix printf specifier handling: invalid specifier should not be marked as "consuming data arguments" Reviewers: rsmith, bruno, dexonsmith Subscribers: cfe-commits Differential Revision: https://reviews.llvm.org/D27796 llvm-svn: 289850
2016-12-16 02:54:00 +08:00
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK: store volatile i32 8, i32* %[[LEN]], align 4
Fix printf specifier handling: invalid specifier should not be marked as "consuming data arguments" Reviewers: rsmith, bruno, dexonsmith Subscribers: cfe-commits Differential Revision: https://reviews.llvm.org/D27796 llvm-svn: 289850
2016-12-16 02:54:00 +08:00
len = __builtin_os_log_format_buffer_size("invalid specifier %: %d even a trailing one%", data);
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK: %[[V1:.*]] = load i8*, i8** %[[BUF_ADDR]], align 8
// CHECK: %[[V2:.*]] = load i32, i32* %[[DATA_ADDR]], align 4
// CHECK: call void @__os_log_helper_1_0_1_4_0(i8* %[[V1]], i32 %[[V2]])
Fix printf specifier handling: invalid specifier should not be marked as "consuming data arguments" Reviewers: rsmith, bruno, dexonsmith Subscribers: cfe-commits Differential Revision: https://reviews.llvm.org/D27796 llvm-svn: 289850
2016-12-16 02:54:00 +08:00
__builtin_os_log_format(buf, "invalid specifier %: %d even a trailing one%", data);
}
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK-LABEL: define linkonce_odr hidden void @__os_log_helper_1_0_1_4_0
// CHECK: (i8* %[[BUFFER:.*]], i32 %[[ARG0:.*]])
// CHECK: %[[BUFFER_ADDR:.*]] = alloca i8*, align 8
// CHECK: %[[ARG0_ADDR:.*]] = alloca i32, align 4
// CHECK: store i8* %[[BUFFER]], i8** %[[BUFFER_ADDR]], align 8
// CHECK: store i32 %[[ARG0]], i32* %[[ARG0_ADDR]], align 4
// CHECK: %[[BUF:.*]] = load i8*, i8** %[[BUFFER_ADDR]], align 8
// CHECK: %[[SUMMARY:.*]] = getelementptr i8, i8* %[[BUF]], i64 0
// CHECK: store i8 0, i8* %[[SUMMARY]], align 1
// CHECK: %[[NUMARGS:.*]] = getelementptr i8, i8* %[[BUF]], i64 1
// CHECK: store i8 1, i8* %[[NUMARGS]], align 1
// CHECK: %[[ARGDESCRIPTOR:.*]] = getelementptr i8, i8* %[[BUF]], i64 2
// CHECK: store i8 0, i8* %[[ARGDESCRIPTOR]], align 1
// CHECK: %[[ARGSIZE:.*]] = getelementptr i8, i8* %[[BUF]], i64 3
// CHECK: store i8 4, i8* %[[ARGSIZE]], align 1
// CHECK: %[[ARGDATA:.*]] = getelementptr i8, i8* %[[BUF]], i64 4
// CHECK: %[[ARGDATACAST:.*]] = bitcast i8* %[[ARGDATA]] to i32*
// CHECK: %[[V0:.*]] = load i32, i32* %[[ARG0_ADDR]], align 4
// CHECK: store i32 %[[V0]], i32* %[[ARGDATACAST]], align 1
Add support for __builtin_os_log_format[_buffer_size] This reverts commit r285007 and reapply r284990, with a fix for the opencl test that I broke. Original commit message follows: These new builtins support a mechanism for logging OS events, using a printf-like format string to specify the layout of data in a buffer. The _buffer_size version of the builtin can be used to determine the size of the buffer to allocate to hold the data, and then __builtin_os_log_format can write data into that buffer. This implements format checking to report mismatches between the format string and the data arguments. Most of this code was written by Chris Willmore. Differential Revision: https://reviews.llvm.org/D25888 llvm-svn: 285019
2016-10-25 04:39:34 +08:00
// CHECK-LABEL: define void @test_builtin_os_log_percent
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK: (i8* %[[BUF:.*]], i8* %[[DATA1:.*]], i8* %[[DATA2:.*]])
Add support for __builtin_os_log_format[_buffer_size] This reverts commit r285007 and reapply r284990, with a fix for the opencl test that I broke. Original commit message follows: These new builtins support a mechanism for logging OS events, using a printf-like format string to specify the layout of data in a buffer. The _buffer_size version of the builtin can be used to determine the size of the buffer to allocate to hold the data, and then __builtin_os_log_format can write data into that buffer. This implements format checking to report mismatches between the format string and the data arguments. Most of this code was written by Chris Willmore. Differential Revision: https://reviews.llvm.org/D25888 llvm-svn: 285019
2016-10-25 04:39:34 +08:00
// Check that the %% which does not consume any argument is correctly handled
Fix handling of %% format specifier in os_log builtins. Returning `false` was stopping the parsing of further arguments, which wasn't intended. llvm-svn: 285047
2016-10-25 08:48:48 +08:00
void test_builtin_os_log_percent(void *buf, const char *data1, const char *data2) {
Add support for __builtin_os_log_format[_buffer_size] This reverts commit r285007 and reapply r284990, with a fix for the opencl test that I broke. Original commit message follows: These new builtins support a mechanism for logging OS events, using a printf-like format string to specify the layout of data in a buffer. The _buffer_size version of the builtin can be used to determine the size of the buffer to allocate to hold the data, and then __builtin_os_log_format can write data into that buffer. This implements format checking to report mismatches between the format string and the data arguments. Most of this code was written by Chris Willmore. Differential Revision: https://reviews.llvm.org/D25888 llvm-svn: 285019
2016-10-25 04:39:34 +08:00
volatile int len;
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK: %[[BUF_ADDR:.*]] = alloca i8*, align 8
// CHECK: %[[DATA1_ADDR:.*]] = alloca i8*, align 8
// CHECK: %[[DATA2_ADDR:.*]] = alloca i8*, align 8
// CHECK: %[[LEN:.*]] = alloca i32, align 4
// CHECK: store i8* %[[BUF]], i8** %[[BUF_ADDR]], align 8
// CHECK: store i8* %[[DATA1]], i8** %[[DATA1_ADDR]], align 8
// CHECK: store i8* %[[DATA2]], i8** %[[DATA2_ADDR]], align 8
// CHECK: store volatile i32 22, i32* %[[LEN]], align 4
Fix handling of %% format specifier in os_log builtins. Returning `false` was stopping the parsing of further arguments, which wasn't intended. llvm-svn: 285047
2016-10-25 08:48:48 +08:00
len = __builtin_os_log_format_buffer_size("%s %% %s", data1, data2);
Add support for __builtin_os_log_format[_buffer_size] This reverts commit r285007 and reapply r284990, with a fix for the opencl test that I broke. Original commit message follows: These new builtins support a mechanism for logging OS events, using a printf-like format string to specify the layout of data in a buffer. The _buffer_size version of the builtin can be used to determine the size of the buffer to allocate to hold the data, and then __builtin_os_log_format can write data into that buffer. This implements format checking to report mismatches between the format string and the data arguments. Most of this code was written by Chris Willmore. Differential Revision: https://reviews.llvm.org/D25888 llvm-svn: 285019
2016-10-25 04:39:34 +08:00
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK: %[[V1:.*]] = load i8*, i8** %[[BUF_ADDR]], align 8
// CHECK: %[[V2:.*]] = load i8*, i8** %[[DATA1_ADDR]], align 8
// CHECK: %[[V3:.*]] = ptrtoint i8* %[[V2]] to i64
// CHECK: %[[V4:.*]] = load i8*, i8** %[[DATA2_ADDR]], align 8
// CHECK: %[[V5:.*]] = ptrtoint i8* %[[V4]] to i64
// CHECK: call void @__os_log_helper_1_2_2_8_32_8_32(i8* %[[V1]], i64 %[[V3]], i64 %[[V5]])
Fix handling of %% format specifier in os_log builtins. Returning `false` was stopping the parsing of further arguments, which wasn't intended. llvm-svn: 285047
2016-10-25 08:48:48 +08:00
__builtin_os_log_format(buf, "%s %% %s", data1, data2);
Add support for __builtin_os_log_format[_buffer_size] This reverts commit r285007 and reapply r284990, with a fix for the opencl test that I broke. Original commit message follows: These new builtins support a mechanism for logging OS events, using a printf-like format string to specify the layout of data in a buffer. The _buffer_size version of the builtin can be used to determine the size of the buffer to allocate to hold the data, and then __builtin_os_log_format can write data into that buffer. This implements format checking to report mismatches between the format string and the data arguments. Most of this code was written by Chris Willmore. Differential Revision: https://reviews.llvm.org/D25888 llvm-svn: 285019
2016-10-25 04:39:34 +08:00
}
Fix test on non-X86 platforms This is a fixup for r285019, adding an `#ifdef __x86_64__` since the os_log builtin is platform specific. llvm-svn: 285027
2016-10-25 05:22:01 +08:00
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK-LABEL: define linkonce_odr hidden void @__os_log_helper_1_2_2_8_32_8_32
// CHECK: (i8* %[[BUFFER:.*]], i64 %[[ARG0:.*]], i64 %[[ARG1:.*]])
// CHECK: %[[BUFFER_ADDR:.*]] = alloca i8*, align 8
// CHECK: %[[ARG0_ADDR:.*]] = alloca i64, align 8
// CHECK: %[[ARG1_ADDR:.*]] = alloca i64, align 8
// CHECK: store i8* %[[BUFFER]], i8** %[[BUFFER_ADDR]], align 8
// CHECK: store i64 %[[ARG0]], i64* %[[ARG0_ADDR]], align 8
// CHECK: store i64 %[[ARG1]], i64* %[[ARG1_ADDR]], align 8
// CHECK: %[[BUF:.*]] = load i8*, i8** %[[BUFFER_ADDR]], align 8
// CHECK: %[[SUMMARY:.*]] = getelementptr i8, i8* %[[BUF]], i64 0
// CHECK: store i8 2, i8* %[[SUMMARY]], align 1
// CHECK: %[[NUMARGS:.*]] = getelementptr i8, i8* %[[BUF]], i64 1
// CHECK: store i8 2, i8* %[[NUMARGS]], align 1
// CHECK: %[[ARGDESCRIPTOR:.*]] = getelementptr i8, i8* %[[BUF]], i64 2
// CHECK: store i8 32, i8* %[[ARGDESCRIPTOR]], align 1
// CHECK: %[[ARGSIZE:.*]] = getelementptr i8, i8* %[[BUF]], i64 3
// CHECK: store i8 8, i8* %[[ARGSIZE]], align 1
// CHECK: %[[ARGDATA:.*]] = getelementptr i8, i8* %[[BUF]], i64 4
// CHECK: %[[ARGDATACAST:.*]] = bitcast i8* %[[ARGDATA]] to i64*
// CHECK: %[[V0:.*]] = load i64, i64* %[[ARG0_ADDR]], align 8
// CHECK: store i64 %[[V0]], i64* %[[ARGDATACAST]], align 1
// CHECK: %[[ARGDESCRIPTOR1:.*]] = getelementptr i8, i8* %[[BUF]], i64 12
// CHECK: store i8 32, i8* %[[ARGDESCRIPTOR1]], align 1
// CHECK: %[[ARGSIZE2:.*]] = getelementptr i8, i8* %[[BUF]], i64 13
// CHECK: store i8 8, i8* %[[ARGSIZE2]], align 1
// CHECK: %[[ARGDATA3:.*]] = getelementptr i8, i8* %[[BUF]], i64 14
// CHECK: %[[ARGDATACAST4:.*]] = bitcast i8* %[[ARGDATA3]] to i64*
// CHECK: %[[V1:.*]] = load i64, i64* %[[ARG1_ADDR]], align 8
// CHECK: store i64 %[[V1]], i64* %[[ARGDATACAST4]], align 1
// Check that the following two functions call the same helper function.
// CHECK-LABEL: define void @test_builtin_os_log_merge_helper0
// CHECK: call void @__os_log_helper_1_0_2_4_0_8_0(
void test_builtin_os_log_merge_helper0(void *buf, int i, double d) {
__builtin_os_log_format(buf, "%d %f", i, d);
}
// CHECK-LABEL: define linkonce_odr hidden void @__os_log_helper_1_0_2_4_0_8_0(
// CHECK-LABEL: define void @test_builtin_os_log_merge_helper1
// CHECK: call void @__os_log_helper_1_0_2_4_0_8_0(
void test_builtin_os_log_merge_helper1(void *buf, unsigned u, long long ll) {
__builtin_os_log_format(buf, "%u %lld", u, ll);
}
// Check that this function doesn't write past the end of array 'buf'.
// CHECK-LABEL: define void @test_builtin_os_log_errno
void test_builtin_os_log_errno() {
Reapply Logging: make os_log buffer size an integer constant expression. The size of an os_log buffer is known at any stage of compilation, so making it a constant expression means that the common idiom of declaring a buffer for it won't result in a VLA. That allows the compiler to skip saving and restoring the stack pointer around such buffers. This also moves the OSLog and other FormatString helpers from libclangAnalysis to libclangAST to avoid a circular dependency. llvm-svn: 345971
2018-11-02 21:14:11 +08:00
// CHECK-NOT: @stacksave
// CHECK: %[[BUF:.*]] = alloca [4 x i8], align 1
// CHECK: %[[DECAY:.*]] = getelementptr inbounds [4 x i8], [4 x i8]* %[[BUF]], i32 0, i32 0
// CHECK: call void @__os_log_helper_1_2_1_0_96(i8* %[[DECAY]])
// CHECK-NOT: @stackrestore
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
char buf[__builtin_os_log_format_buffer_size("%m")];
__builtin_os_log_format(buf, "%m");
}
Reapply Logging: make os_log buffer size an integer constant expression. The size of an os_log buffer is known at any stage of compilation, so making it a constant expression means that the common idiom of declaring a buffer for it won't result in a VLA. That allows the compiler to skip saving and restoring the stack pointer around such buffers. This also moves the OSLog and other FormatString helpers from libclangAnalysis to libclangAST to avoid a circular dependency. llvm-svn: 345971
2018-11-02 21:14:11 +08:00
// CHECK-LABEL: define linkonce_odr hidden void @__os_log_helper_1_2_1_0_96
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK: (i8* %[[BUFFER:.*]])
// CHECK: %[[BUFFER_ADDR:.*]] = alloca i8*, align 8
// CHECK: store i8* %[[BUFFER]], i8** %[[BUFFER_ADDR]], align 8
// CHECK: %[[BUF:.*]] = load i8*, i8** %[[BUFFER_ADDR]], align 8
// CHECK: %[[SUMMARY:.*]] = getelementptr i8, i8* %[[BUF]], i64 0
Reapply Logging: make os_log buffer size an integer constant expression. The size of an os_log buffer is known at any stage of compilation, so making it a constant expression means that the common idiom of declaring a buffer for it won't result in a VLA. That allows the compiler to skip saving and restoring the stack pointer around such buffers. This also moves the OSLog and other FormatString helpers from libclangAnalysis to libclangAST to avoid a circular dependency. llvm-svn: 345971
2018-11-02 21:14:11 +08:00
// CHECK: store i8 2, i8* %[[SUMMARY]], align 1
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK: %[[NUMARGS:.*]] = getelementptr i8, i8* %[[BUF]], i64 1
// CHECK: store i8 1, i8* %[[NUMARGS]], align 1
// CHECK: %[[ARGDESCRIPTOR:.*]] = getelementptr i8, i8* %[[BUF]], i64 2
Reapply Logging: make os_log buffer size an integer constant expression. The size of an os_log buffer is known at any stage of compilation, so making it a constant expression means that the common idiom of declaring a buffer for it won't result in a VLA. That allows the compiler to skip saving and restoring the stack pointer around such buffers. This also moves the OSLog and other FormatString helpers from libclangAnalysis to libclangAST to avoid a circular dependency. llvm-svn: 345971
2018-11-02 21:14:11 +08:00
// CHECK: store i8 96, i8* %[[ARGDESCRIPTOR]], align 1
[CodeGen] Emit a helper function for __builtin_os_log_format to reduce code size. Currently clang expands a call to __builtin_os_log_format into a long sequence of instructions at the call site, causing code size to increase in some cases. This commit attempts to reduce code size by emitting a helper function that can be shared by calls to __builtin_os_log_format with similar formats and arguments. The helper function has linkonce_odr linkage to enable the linker to merge identical functions across translation units. Attribute 'noinline' is attached to the helper function at -Oz so that the inliner doesn't inline functions that can potentially be merged. This commit also fixes a bug where the generated IR writes past the end of the buffer when "%m" is the last specifier appearing in the format string passed to __builtin_os_log_format. Original patch by Duncan Exon Smith. rdar://problem/34065973 rdar://problem/34196543 Differential Revision: https://reviews.llvm.org/D38606 llvm-svn: 315045
2017-10-06 15:12:46 +08:00
// CHECK: %[[ARGSIZE:.*]] = getelementptr i8, i8* %[[BUF]], i64 3
// CHECK: store i8 0, i8* %[[ARGSIZE]], align 1
// CHECK-NEXT: ret void
// CHECK-LABEL: define void @test_builtin_os_log_long_double
// CHECK: (i8* %[[BUF:.*]], x86_fp80 %[[LD:.*]])
void test_builtin_os_log_long_double(void *buf, long double ld) {
// CHECK: %[[BUF_ADDR:.*]] = alloca i8*, align 8
// CHECK: %[[LD_ADDR:.*]] = alloca x86_fp80, align 16
// CHECK: %[[COERCE:.*]] = alloca i128, align 16
// CHECK: store i8* %[[BUF]], i8** %[[BUF_ADDR]], align 8
// CHECK: store x86_fp80 %[[LD]], x86_fp80* %[[LD_ADDR]], align 16
// CHECK: %[[V0:.*]] = load i8*, i8** %[[BUF_ADDR]], align 8
// CHECK: %[[V1:.*]] = load x86_fp80, x86_fp80* %[[LD_ADDR]], align 16
// CHECK: %[[V2:.*]] = bitcast x86_fp80 %[[V1]] to i80
// CHECK: %[[V3:.*]] = zext i80 %[[V2]] to i128
// CHECK: store i128 %[[V3]], i128* %[[COERCE]], align 16
// CHECK: %[[V4:.*]] = bitcast i128* %[[COERCE]] to { i64, i64 }*
// CHECK: %[[V5:.*]] = getelementptr inbounds { i64, i64 }, { i64, i64 }* %[[V4]], i32 0, i32 0
// CHECK: %[[V6:.*]] = load i64, i64* %[[V5]], align 16
// CHECK: %[[V7:.*]] = getelementptr inbounds { i64, i64 }, { i64, i64 }* %[[V4]], i32 0, i32 1
// CHECK: %[[V8:.*]] = load i64, i64* %[[V7]], align 8
// CHECK: call void @__os_log_helper_1_0_1_16_0(i8* %[[V0]], i64 %[[V6]], i64 %[[V8]])
__builtin_os_log_format(buf, "%Lf", ld);
}
// CHECK-LABEL: define linkonce_odr hidden void @__os_log_helper_1_0_1_16_0
// CHECK: (i8* %[[BUFFER:.*]], i64 %[[ARG0_COERCE0:.*]], i64 %[[ARG0_COERCE1:.*]])
// CHECK: %[[ARG0:.*]] = alloca i128, align 16
// CHECK: %[[BUFFER_ADDR:.*]] = alloca i8*, align 8
// CHECK: %[[ARG0_ADDR:.*]] = alloca i128, align 16
// CHECK: %[[V0:.*]] = bitcast i128* %[[ARG0]] to { i64, i64 }*
// CHECK: %[[V1:.*]] = getelementptr inbounds { i64, i64 }, { i64, i64 }* %[[V0]], i32 0, i32 0
// CHECK: store i64 %[[ARG0_COERCE0]], i64* %[[V1]], align 16
// CHECK: %[[V2:.*]] = getelementptr inbounds { i64, i64 }, { i64, i64 }* %[[V0]], i32 0, i32 1
// CHECK: store i64 %[[ARG0_COERCE1]], i64* %[[V2]], align 8
// CHECK: %[[ARG01:.*]] = load i128, i128* %[[ARG0]], align 16
// CHECK: store i8* %[[BUFFER]], i8** %[[BUFFER_ADDR]], align 8
// CHECK: store i128 %[[ARG01]], i128* %[[ARG0_ADDR]], align 16
// CHECK: %[[BUF:.*]] = load i8*, i8** %[[BUFFER_ADDR]], align 8
// CHECK: %[[SUMMARY:.*]] = getelementptr i8, i8* %[[BUF]], i64 0
// CHECK: store i8 0, i8* %[[SUMMARY]], align 1
// CHECK: %[[NUMARGS:.*]] = getelementptr i8, i8* %[[BUF]], i64 1
// CHECK: store i8 1, i8* %[[NUMARGS]], align 1
// CHECK: %[[ARGDESCRIPTOR:.*]] = getelementptr i8, i8* %[[BUF]], i64 2
// CHECK: store i8 0, i8* %[[ARGDESCRIPTOR]], align 1
// CHECK: %[[ARGSIZE:.*]] = getelementptr i8, i8* %[[BUF]], i64 3
// CHECK: store i8 16, i8* %[[ARGSIZE]], align 1
// CHECK: %[[ARGDATA:.*]] = getelementptr i8, i8* %[[BUF]], i64 4
// CHECK: %[[ARGDATACAST:.*]] = bitcast i8* %[[ARGDATA]] to i128*
// CHECK: %[[V3:.*]] = load i128, i128* %[[ARG0_ADDR]], align 16
// CHECK: store i128 %[[V3]], i128* %[[ARGDATACAST]], align 1
#endif