int main(int argc, char *argv[]) {
INPUT_TYPE input[INPUT_SIZE];
int i, j;
srand(42);
// Initialize the input array with data of various sizes.
for (i=0; i<INPUT_SIZE; ++i)
input[i] = rand() & 0x3f;
int64_t fixedInput = INT64_C(0x1234567890ABCDEF);
double bestTime = __builtin_inf();
void *dummyp;
for (j=0; j<1024; ++j) {
uint64_t startTime = mach_absolute_time();
for (i=0; i<INPUT_SIZE; ++i)
FUNCTION_NAME(fixedInput, input[i]);
uint64_t endTime = mach_absolute_time();
double thisTime = intervalInCycles(startTime, endTime);
bestTime = __builtin_fmin(thisTime, bestTime);
// Move the stack alignment between trials to eliminate (mostly) aliasing effects
dummyp = alloca(1);
}
printf("%16s: %f cycles.\n", LIBSTRING, bestTime / (double) INPUT_SIZE);
return 0;
}
double
vector_min (void)
{
size_t i;
double min = values[0];
for (i = 0; i < SIZE; i++)
min = __builtin_fmin (min, values[i]);
return min;
}
inline Y fmin(const Y& v1,const Y& v2)
{
return __builtin_fmin(v1,v2);
}
// 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
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
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
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
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
}