intel/llvm
1
0
Fork 0
mirror of https://github.com/intel/llvm.git synced 2026-01-20 10:58:11 +08:00
Code Issues Packages Projects Releases Wiki Activity

Speculatively fix build bots

Browse Source 
This should address the issues found by:
https://lab.llvm.org/buildbot/#/builders/105/builds/24292
This commit is contained in:
Aaron Ballman
2022-04-20 12:02:16 -04:00
parent 4f843f9a9e
commit 7661d3bc9d
4 changed files with 147 additions and 143 deletions
Download Patch File Download Diff File Expand all files Collapse all files

81
compiler-rt/test/builtins/Unit/ppc/qadd_test.c
View File

@@ -4,6 +4,7 @@
#include "DD.h"
long double __gcc_qadd(long double x, long double y);
int memcmp(const void *, const void *, __typeof__(sizeof(0)));
double testAccuracy();
int testEdgeCases();
@@ -11,10 +12,10 @@ int testEdgeCases();
int main(int argc, char *argv[]) {
if (testEdgeCases())
return 1;
if (testAccuracy() > 1.0)
return 1;
return 0;
}
@@ -114,11 +115,11 @@ int testEdgeCases() {
b.lo = edgeCases[i].ylo;
r.hi = edgeCases[i].rhi;
r.lo = edgeCases[i].rlo;
int error;
DD c = { .ld = __gcc_qadd(a.ld, b.ld) };
if (r.hi != r.hi) {
if (c.hi == c.hi)
error = 1;
@@ -130,16 +131,16 @@ int testEdgeCases() {
else if (r.hi == 0.0)
error = memcmp(&c, &r, sizeof(DD));
else
error = ((c.hi != r.hi) || (c.lo != r.lo));
if (error) {
printf("Error on edge case %a + %a: expected (%a, %a), got (%a, %a).\n", a.hi, b.hi, r.hi, r.lo, c.hi, c.lo);
return 1;
}
}
return 0;
}
@@ -147,117 +148,117 @@ int testEdgeCases() {
/*
Code for generating the test cases, requires the mpfr package to run.
#include <stdio.h>
#include <stdlib.h>
#include <mpfr.h>
#ifdef __x86_64__
#define randlength 2
#else
#define randlength 4
#endif
int main(int argc, char *argv[]) {
MPFR_DECL_INIT(a, 106);
MPFR_DECL_INIT(b, 106);
MPFR_DECL_INIT(c, 106);
MPFR_DECL_INIT(tmp, 53);
int exponent_range = atoi(argv[1]);
int i;
for (i=0; i<128; ++i) {
mpfr_random2(a, randlength, exponent_range);
mpfr_random2(b, randlength, exponent_range);
mpfr_add(c, a, b, GMP_RNDN);
double ahi = mpfr_get_d(a, GMP_RNDN);
mpfr_set_d(tmp, ahi, GMP_RNDN);
mpfr_sub(tmp, a, tmp, GMP_RNDN);
double alo = mpfr_get_d(tmp, GMP_RNDN);
printf("{%0.13a, %0.13a, ", ahi, alo);
double bhi = mpfr_get_d(b, GMP_RNDN);
mpfr_set_d(tmp, bhi, GMP_RNDN);
mpfr_sub(tmp, b, tmp, GMP_RNDN);
double blo = mpfr_get_d(tmp, GMP_RNDN);
printf("%0.13a, %0.13a, ", bhi, blo);
double chi = mpfr_get_d(c, GMP_RNDN);
mpfr_set_d(tmp, chi, GMP_RNDN);
mpfr_sub(tmp, c, tmp, GMP_RNDN);
double clo = mpfr_get_d(tmp, GMP_RNDN);
printf("%0.13a, %0.13a},\n", chi, clo);
mpfr_neg(b, b, GMP_RNDN);
mpfr_add(c, a, b, GMP_RNDN);
ahi = mpfr_get_d(a, GMP_RNDN);
mpfr_set_d(tmp, ahi, GMP_RNDN);
mpfr_sub(tmp, a, tmp, GMP_RNDN);
alo = mpfr_get_d(tmp, GMP_RNDN);
printf("{%0.13a, %0.13a, ", ahi, alo);
bhi = mpfr_get_d(b, GMP_RNDN);
mpfr_set_d(tmp, bhi, GMP_RNDN);
mpfr_sub(tmp, b, tmp, GMP_RNDN);
blo = mpfr_get_d(tmp, GMP_RNDN);
printf("%0.13a, %0.13a, ", bhi, blo);
chi = mpfr_get_d(c, GMP_RNDN);
mpfr_set_d(tmp, chi, GMP_RNDN);
mpfr_sub(tmp, c, tmp, GMP_RNDN);
clo = mpfr_get_d(tmp, GMP_RNDN);
printf("%0.13a, %0.13a},\n", chi, clo);
mpfr_neg(a, a, GMP_RNDN);
mpfr_neg(b, b, GMP_RNDN);
mpfr_add(c, a, b, GMP_RNDN);
ahi = mpfr_get_d(a, GMP_RNDN);
mpfr_set_d(tmp, ahi, GMP_RNDN);
mpfr_sub(tmp, a, tmp, GMP_RNDN);
alo = mpfr_get_d(tmp, GMP_RNDN);
printf("{%0.13a, %0.13a, ", ahi, alo);
bhi = mpfr_get_d(b, GMP_RNDN);
mpfr_set_d(tmp, bhi, GMP_RNDN);
mpfr_sub(tmp, b, tmp, GMP_RNDN);
blo = mpfr_get_d(tmp, GMP_RNDN);
printf("%0.13a, %0.13a, ", bhi, blo);
chi = mpfr_get_d(c, GMP_RNDN);
mpfr_set_d(tmp, chi, GMP_RNDN);
mpfr_sub(tmp, c, tmp, GMP_RNDN);
clo = mpfr_get_d(tmp, GMP_RNDN);
printf("%0.13a, %0.13a},\n", chi, clo);
mpfr_neg(b, b, GMP_RNDN);
mpfr_add(c, a, b, GMP_RNDN);
ahi = mpfr_get_d(a, GMP_RNDN);
mpfr_set_d(tmp, ahi, GMP_RNDN);
mpfr_sub(tmp, a, tmp, GMP_RNDN);
alo = mpfr_get_d(tmp, GMP_RNDN);
printf("{%0.13a, %0.13a, ", ahi, alo);
bhi = mpfr_get_d(b, GMP_RNDN);
mpfr_set_d(tmp, bhi, GMP_RNDN);
mpfr_sub(tmp, b, tmp, GMP_RNDN);
blo = mpfr_get_d(tmp, GMP_RNDN);
printf("%0.13a, %0.13a, ", bhi, blo);
chi = mpfr_get_d(c, GMP_RNDN);
mpfr_set_d(tmp, chi, GMP_RNDN);
mpfr_sub(tmp, c, tmp, GMP_RNDN);
clo = mpfr_get_d(tmp, GMP_RNDN);
printf("%0.13a, %0.13a},\n", chi, clo);
}
return 0;
}
@@ -1813,9 +1814,9 @@ const int numAccuracyTests = sizeof(accuracyTests) / sizeof(struct testVector);
double testAccuracy() {
int i;
DD a, b, c, r;
double worstUlps = 0.5;
for (i=0; i<numAccuracyTests; ++i) {
a.hi = accuracyTests[i].xhi;
a.lo = accuracyTests[i].xlo;
@@ -1823,17 +1824,17 @@ double testAccuracy() {
b.lo = accuracyTests[i].ylo;
r.hi = accuracyTests[i].rhi;
r.lo = accuracyTests[i].rlo;
DD c = { .ld = __gcc_qadd(a.ld, b.ld) };
double error = __builtin_fabs(((r.hi - c.hi) + r.lo) - c.lo);
if (error != 0.0) {
int exponent = ilogb(r.hi);
exponent = (exponent < -1022 ? -1022 : exponent);
double ulpError = scalbn(error, 106 - exponent);
if (ulpError > worstUlps) {
#ifdef PRINT_ACCURACY_INFORMATION
printf("New worst rounding error for (%a,%a) + (%a,%a):\n", a.hi, a.lo, b.hi, b.lo);
@@ -1845,7 +1846,7 @@ double testAccuracy() {
}
}
}
return worstUlps;
}

63
compiler-rt/test/builtins/Unit/ppc/qdiv_test.c
View File

@@ -4,6 +4,7 @@
#include "DD.h"
long double __gcc_qdiv(long double x, long double y);
int memcmp(const void *, const void *, __typeof__(sizeof(0)));
double testAccuracy();
int testEdgeCases();
@@ -11,10 +12,10 @@ int testEdgeCases();
int main(int argc, char *argv[]) {
if (testEdgeCases())
return 1;
if (testAccuracy() > 4.0)
return 1;
return 0;
}
@@ -114,11 +115,11 @@ int testEdgeCases() {
b.lo = edgeCases[i].ylo;
r.hi = edgeCases[i].rhi;
r.lo = edgeCases[i].rlo;
int error;
DD c = { .ld = __gcc_qdiv(a.ld, b.ld) };
if (r.hi != r.hi) {
if (c.hi == c.hi)
error = 1;
@@ -130,16 +131,16 @@ int testEdgeCases() {
else if (r.hi == 0.0)
error = memcmp(&c, &r, sizeof(DD));
else
error = ((c.hi != r.hi) || (c.lo != r.lo));
if (error) {
printf("Error on edge case %a / %a: expected (%a, %a), got (%a, %a).\n", a.hi, b.hi, r.hi, r.lo, c.hi, c.lo);
return 1;
}
}
return 0;
}
@@ -150,68 +151,68 @@ int testEdgeCases() {
#include <stdlib.h>
#include <mpfr.h>
#include <math.h>
#ifdef __x86_64__
#define randlength 2
#else
#define randlength 4
#endif
void printTest(mpfr_t a, mpfr_t b, mpfr_t c) {
static const double infinityD = __builtin_inf();
MPFR_DECL_INIT(tmp, 53);
double ahi = mpfr_get_d(a, GMP_RNDN);
mpfr_set_d(tmp, ahi, GMP_RNDN);
mpfr_sub(tmp, a, tmp, GMP_RNDN);
double alo = mpfr_get_d(tmp, GMP_RNDN);
printf("{%0.13a, %0.13a, ", ahi, alo);
double bhi = mpfr_get_d(b, GMP_RNDN);
mpfr_set_d(tmp, bhi, GMP_RNDN);
mpfr_sub(tmp, b, tmp, GMP_RNDN);
double blo = mpfr_get_d(tmp, GMP_RNDN);
printf("%0.13a, %0.13a, ", bhi, blo);
double chi = mpfr_get_d(c, GMP_RNDN);
mpfr_set_d(tmp, chi, GMP_RNDN);
mpfr_sub(tmp, c, tmp, GMP_RNDN);
double clo = isinf(chi) ? 0.0 : mpfr_get_d(tmp, GMP_RNDN);
printf("%0.13a, %0.13a},\n", chi, clo);
}
int main(int argc, char *argv[]) {
MPFR_DECL_INIT(a, 106);
MPFR_DECL_INIT(b, 106);
MPFR_DECL_INIT(c, 106);
int exponent_range = atoi(argv[1]);
int i;
for (i=0; i<128; ++i) {
mpfr_random2(a, randlength, exponent_range);
mpfr_random2(b, randlength, exponent_range);
mpfr_div(c, a, b, GMP_RNDN);
printTest(a, b, c);
mpfr_neg(b, b, GMP_RNDN);
mpfr_div(c, a, b, GMP_RNDN);
printTest(a, b, c);
mpfr_neg(a, a, GMP_RNDN);
mpfr_neg(b, b, GMP_RNDN);
mpfr_div(c, a, b, GMP_RNDN);
printTest(a, b, c);
mpfr_neg(b, b, GMP_RNDN);
mpfr_div(c, a, b, GMP_RNDN);
printTest(a, b, c);
}
return 0;
}
}
*/
const struct testVector accuracyTests[] = {
@@ -736,9 +737,9 @@ const int numAccuracyTests = sizeof(accuracyTests) / sizeof(struct testVector);
double testAccuracy() {
int i;
DD a, b, c, r;
double worstUlps = 0.5;
for (i=0; i<numAccuracyTests; ++i) {
a.hi = accuracyTests[i].xhi;
a.lo = accuracyTests[i].xlo;
@@ -746,17 +747,17 @@ double testAccuracy() {
b.lo = accuracyTests[i].ylo;
r.hi = accuracyTests[i].rhi;
r.lo = accuracyTests[i].rlo;
DD c = { .ld = __gcc_qdiv(a.ld, b.ld) };
double error = __builtin_fabs(((r.hi - c.hi) + r.lo) - c.lo);
if (error != 0.0) {
int exponent = ilogb(r.hi);
exponent = (exponent < -1022 ? -1022 : exponent);
double ulpError = scalbn(error, 106 - exponent);
if (ulpError > worstUlps) {
#ifdef PRINT_ACCURACY_INFORMATION
printf("New worst rounding error for (%a,%a) / (%a,%a):\n", a.hi, a.lo, b.hi, b.lo);
@@ -768,7 +769,7 @@ double testAccuracy() {
}
}
}
return worstUlps;
}

63
compiler-rt/test/builtins/Unit/ppc/qmul_test.c
View File

@@ -4,6 +4,7 @@
#include "DD.h"
long double __gcc_qmul(long double x, long double y);
int memcmp(const void *, const void *, __typeof__(sizeof(0)));
double testAccuracy();
int testEdgeCases();
@@ -11,10 +12,10 @@ int testEdgeCases();
int main(int argc, char *argv[]) {
if (testEdgeCases())
return 1;
if (testAccuracy() > 2.0)
return 1;
return 0;
}
@@ -114,11 +115,11 @@ int testEdgeCases() {
b.lo = edgeCases[i].ylo;
r.hi = edgeCases[i].rhi;
r.lo = edgeCases[i].rlo;
int error;
DD c = { .ld = __gcc_qmul(a.ld, b.ld) };
if (r.hi != r.hi) {
if (c.hi == c.hi)
error = 1;
@@ -130,16 +131,16 @@ int testEdgeCases() {
else if (r.hi == 0.0)
error = memcmp(&c, &r, sizeof(DD));
else
error = ((c.hi != r.hi) || (c.lo != r.lo));
if (error) {
printf("Error on edge case %a x %a: expected (%a, %a), got (%a, %a).\n", a.hi, b.hi, r.hi, r.lo, c.hi, c.lo);
return 1;
}
}
return 0;
}
@@ -147,72 +148,72 @@ int testEdgeCases() {
/*
Code for generating the test cases, requires the mpfr package to run.
#include <stdio.h>
#include <stdlib.h>
#include <mpfr.h>
#include <math.h>
#ifdef __x86_64__
#define randlength 2
#else
#define randlength 4
#endif
void printTest(mpfr_t a, mpfr_t b, mpfr_t c) {
static const double INFINITYD = __builtin_INFINITY();
MPFR_DECL_INIT(tmp, 53);
double ahi = mpfr_get_d(a, GMP_RNDN);
mpfr_set_d(tmp, ahi, GMP_RNDN);
mpfr_sub(tmp, a, tmp, GMP_RNDN);
double alo = mpfr_get_d(tmp, GMP_RNDN);
printf("{%0.13a, %0.13a, ", ahi, alo);
double bhi = mpfr_get_d(b, GMP_RNDN);
mpfr_set_d(tmp, bhi, GMP_RNDN);
mpfr_sub(tmp, b, tmp, GMP_RNDN);
double blo = mpfr_get_d(tmp, GMP_RNDN);
printf("%0.13a, %0.13a, ", bhi, blo);
double chi = mpfr_get_d(c, GMP_RNDN);
mpfr_set_d(tmp, chi, GMP_RNDN);
mpfr_sub(tmp, c, tmp, GMP_RNDN);
double clo = isINFINITY(chi) ? 0.0 : mpfr_get_d(tmp, GMP_RNDN);
printf("%0.13a, %0.13a},\n", chi, clo);
}
int main(int argc, char *argv[]) {
MPFR_DECL_INIT(a, 106);
MPFR_DECL_INIT(b, 106);
MPFR_DECL_INIT(c, 106);
int exponent_range = atoi(argv[1]);
int i;
for (i=0; i<128; ++i) {
mpfr_random2(a, randlength, exponent_range);
mpfr_random2(b, randlength, exponent_range);
mpfr_mul(c, a, b, GMP_RNDN);
printTest(a, b, c);
mpfr_neg(b, b, GMP_RNDN);
mpfr_mul(c, a, b, GMP_RNDN);
printTest(a, b, c);
mpfr_neg(a, a, GMP_RNDN);
mpfr_neg(b, b, GMP_RNDN);
mpfr_mul(c, a, b, GMP_RNDN);
printTest(a, b, c);
mpfr_neg(b, b, GMP_RNDN);
mpfr_mul(c, a, b, GMP_RNDN);
printTest(a, b, c);
}
return 0;
}
}
*/
@@ -734,9 +735,9 @@ const int numAccuracyTests = sizeof(accuracyTests) / sizeof(struct testVector);
double testAccuracy() {
int i;
DD a, b, c, r;
double worstUlps = 0.5;
for (i=0; i<numAccuracyTests; ++i) {
a.hi = accuracyTests[i].xhi;
a.lo = accuracyTests[i].xlo;
@@ -744,17 +745,17 @@ double testAccuracy() {
b.lo = accuracyTests[i].ylo;
r.hi = accuracyTests[i].rhi;
r.lo = accuracyTests[i].rlo;
DD c = { .ld = __gcc_qmul(a.ld, b.ld) };
double error = __builtin_fabs(((r.hi - c.hi) + r.lo) - c.lo);
if (error != 0.0) {
int exponent = ilogb(r.hi);
exponent = (exponent < -1022 ? -1022 : exponent);
double ulpError = scalbn(error, 106 - exponent);
if (ulpError > worstUlps) {
#ifdef PRINT_ACCURACY_INFORMATION
printf("New worst rounding error for (%a,%a) x (%a,%a):\n", a.hi, a.lo, b.hi, b.lo);
@@ -766,7 +767,7 @@ double testAccuracy() {
}
}
}
return worstUlps;
}

83
compiler-rt/test/builtins/Unit/ppc/qsub_test.c
View File

@@ -4,6 +4,7 @@
#include "DD.h"
long double __gcc_qsub(long double x, long double y);
int memcmp(const void *, const void *, __typeof__(sizeof(0)));
double testAccuracy();
int testEdgeCases();
@@ -11,10 +12,10 @@ int testEdgeCases();
int main(int argc, char *argv[]) {
if (testEdgeCases())
return 1;
if (testAccuracy() > 1.0)
return 1;
return 0;
}
@@ -114,13 +115,13 @@ int testEdgeCases() {
b.lo = edgeCases[i].ylo;
r.hi = edgeCases[i].rhi;
r.lo = edgeCases[i].rlo;
int error;
long double computed = __gcc_qsub(a.ld, b.ld);
DD c = { .ld = computed };
if (r.hi != r.hi) {
if (c.hi == c.hi)
error = 1;
@@ -132,16 +133,16 @@ int testEdgeCases() {
else if (r.hi == 0.0)
error = memcmp(&c, &r, sizeof(DD));
else
error = ((c.hi != r.hi) || (c.lo != r.lo));
if (error) {
printf("Error on edge case %a - %a: expected (%a, %a), got (%a, %a).\n", a.hi, b.hi, r.hi, r.lo, c.hi, c.lo);
return 1;
}
}
return 0;
}
@@ -149,117 +150,117 @@ int testEdgeCases() {
/*
Code for generating the test cases, requires the mpfr package to run.
#include <stdio.h>
#include <stdlib.h>
#include <mpfr.h>
#ifdef __x86_64__
#define randlength 2
#else
#define randlength 4
#endif
int main(int argc, char *argv[]) {
MPFR_DECL_INIT(a, 106);
MPFR_DECL_INIT(b, 106);
MPFR_DECL_INIT(c, 106);
MPFR_DECL_INIT(tmp, 53);
int exponent_range = atoi(argv[1]);
int i;
for (i=0; i<128; ++i) {
mpfr_random2(a, randlength, exponent_range);
mpfr_random2(b, randlength, exponent_range);
mpfr_add(c, a, b, GMP_RNDN);
double ahi = mpfr_get_d(a, GMP_RNDN);
mpfr_set_d(tmp, ahi, GMP_RNDN);
mpfr_sub(tmp, a, tmp, GMP_RNDN);
double alo = mpfr_get_d(tmp, GMP_RNDN);
printf("{%0.13a, %0.13a, ", ahi, alo);
double bhi = mpfr_get_d(b, GMP_RNDN);
mpfr_set_d(tmp, bhi, GMP_RNDN);
mpfr_sub(tmp, b, tmp, GMP_RNDN);
double blo = mpfr_get_d(tmp, GMP_RNDN);
printf("%0.13a, %0.13a, ", bhi, blo);
double chi = mpfr_get_d(c, GMP_RNDN);
mpfr_set_d(tmp, chi, GMP_RNDN);
mpfr_sub(tmp, c, tmp, GMP_RNDN);
double clo = mpfr_get_d(tmp, GMP_RNDN);
printf("%0.13a, %0.13a},\n", chi, clo);
mpfr_neg(b, b, GMP_RNDN);
mpfr_add(c, a, b, GMP_RNDN);
ahi = mpfr_get_d(a, GMP_RNDN);
mpfr_set_d(tmp, ahi, GMP_RNDN);
mpfr_sub(tmp, a, tmp, GMP_RNDN);
alo = mpfr_get_d(tmp, GMP_RNDN);
printf("{%0.13a, %0.13a, ", ahi, alo);
bhi = mpfr_get_d(b, GMP_RNDN);
mpfr_set_d(tmp, bhi, GMP_RNDN);
mpfr_sub(tmp, b, tmp, GMP_RNDN);
blo = mpfr_get_d(tmp, GMP_RNDN);
printf("%0.13a, %0.13a, ", bhi, blo);
chi = mpfr_get_d(c, GMP_RNDN);
mpfr_set_d(tmp, chi, GMP_RNDN);
mpfr_sub(tmp, c, tmp, GMP_RNDN);
clo = mpfr_get_d(tmp, GMP_RNDN);
printf("%0.13a, %0.13a},\n", chi, clo);
mpfr_neg(a, a, GMP_RNDN);
mpfr_neg(b, b, GMP_RNDN);
mpfr_add(c, a, b, GMP_RNDN);
ahi = mpfr_get_d(a, GMP_RNDN);
mpfr_set_d(tmp, ahi, GMP_RNDN);
mpfr_sub(tmp, a, tmp, GMP_RNDN);
alo = mpfr_get_d(tmp, GMP_RNDN);
printf("{%0.13a, %0.13a, ", ahi, alo);
bhi = mpfr_get_d(b, GMP_RNDN);
mpfr_set_d(tmp, bhi, GMP_RNDN);
mpfr_sub(tmp, b, tmp, GMP_RNDN);
blo = mpfr_get_d(tmp, GMP_RNDN);
printf("%0.13a, %0.13a, ", bhi, blo);
chi = mpfr_get_d(c, GMP_RNDN);
mpfr_set_d(tmp, chi, GMP_RNDN);
mpfr_sub(tmp, c, tmp, GMP_RNDN);
clo = mpfr_get_d(tmp, GMP_RNDN);
printf("%0.13a, %0.13a},\n", chi, clo);
mpfr_neg(b, b, GMP_RNDN);
mpfr_add(c, a, b, GMP_RNDN);
ahi = mpfr_get_d(a, GMP_RNDN);
mpfr_set_d(tmp, ahi, GMP_RNDN);
mpfr_sub(tmp, a, tmp, GMP_RNDN);
alo = mpfr_get_d(tmp, GMP_RNDN);
printf("{%0.13a, %0.13a, ", ahi, alo);
bhi = mpfr_get_d(b, GMP_RNDN);
mpfr_set_d(tmp, bhi, GMP_RNDN);
mpfr_sub(tmp, b, tmp, GMP_RNDN);
blo = mpfr_get_d(tmp, GMP_RNDN);
printf("%0.13a, %0.13a, ", bhi, blo);
chi = mpfr_get_d(c, GMP_RNDN);
mpfr_set_d(tmp, chi, GMP_RNDN);
mpfr_sub(tmp, c, tmp, GMP_RNDN);
clo = mpfr_get_d(tmp, GMP_RNDN);
printf("%0.13a, %0.13a},\n", chi, clo);
}
return 0;
}
@@ -1815,9 +1816,9 @@ const int numAccuracyTests = sizeof(accuracyTests) / sizeof(struct testVector);
double testAccuracy() {
int i;
DD a, b, c, r;
double worstUlps = 0.5;
for (i=0; i<numAccuracyTests; ++i) {
a.hi = accuracyTests[i].xhi;
a.lo = accuracyTests[i].xlo;
@@ -1825,17 +1826,17 @@ double testAccuracy() {
b.lo = -accuracyTests[i].ylo;
r.hi = accuracyTests[i].rhi;
r.lo = accuracyTests[i].rlo;
DD c = { .ld = __gcc_qsub(a.ld, b.ld) };
double error = __builtin_fabs(((r.hi - c.hi) + r.lo) - c.lo);
if (error != 0.0) {
int exponent = ilogb(r.hi);
exponent = (exponent < -1022 ? -1022 : exponent);
double ulpError = scalbn(error, 106 - exponent);
if (ulpError > worstUlps) {
#ifdef PRINT_ACCURACY_INFORMATION
printf("New worst rounding error for (%a,%a) + (%a,%a):\n", a.hi, a.lo, b.hi, b.lo);
@@ -1847,7 +1848,7 @@ double testAccuracy() {
}
}
}
return worstUlps;
}