Relax gen-auto-libm-tests may-underflow rules, test log1p in all rounding modes
diff mbox

Message ID Pine.LNX.4.64.1403241609520.6513@digraph.polyomino.org.uk
State New
Headers show

Commit Message

Joseph Myers March 24, 2014, 4:10 p.m. UTC
gen-auto-libm-tests presently allows but does not require underflow
exceptions for results with magnitude in the range (greatest
subnormal, least normal].

In some cases, the magnitude of the exact result is very slightly
above the least normal, but rounding in the implementation results in
it effectively computing an infinite-precision result that is slightly
below the least normal, so raising an underflow exception.  This is in
accordance with the documented accuracy goals, but results in
testsuite failures.

This patch changes the logic to allow underflows when the mathematical
result is up to 0.5ulp above the least normal (so in any case where
the round-to-nearest result is the least normal).  Ideally underflows
in all these cases would be accepted only when an underflow with the
actual result is consistent with the rounding mode (in FE_TOWARDZERO
mode, a return value of the least normal implies that the
infinite-precision result did not underflow so there should be no
underflow exception, for example), so as to match the documented goals
more precisely - whereas at present the tests for exceptions are
completely independent of the tests of the returned values.  (The same
applies to overflow exceptions as well - they too should be checked
for consistency with the result, as in FE_TOWARDZERO mode a result
1ulp below the largest finite value should be inconsistent with an
overflow exception and cause a failure with overflow rather than
simply being considered a 1ulp error when overflow is expected.)  But
the present patch at least deals with the cases causing spurious
failures so that (a) certain existing tests no longer need to be
marked as having spurious exceptions (such markings in
auto-libm-test-in end up applying to more cases than just those they
are needed for) and (b) log1p can be tested in all rounding modes
without introducing more such failures.  This patch duly moves tests
of log1p to ALL_RM_TEST.

Tested x86_64 and x86 and ulps updated accordingly.

(auto-libm-test-out changes omitted below.)

2014-03-24  Joseph Myers  <joseph@codesourcery.com>

	[BZ #16357]
	[BZ #16599]
	* math/gen-auto-libm-tests.c (fp_format_desc): Add field
	min_plus_half.
	(fp_formats): Update initializers.
	(init_fp_formats): Initialize new field.
	(output_for_one_input_case): Allow underflow for results up to
	min_plus_half.
	* math/libm-test.inc (log1p_test): Use ALL_RM_TEST.
	* math/auto-libm-test-in: Don't mark some underflows from asin and
	atanh as spurious.
	* math/auto-libm-test-out: Regenerated.
	* sysdeps/i386/fpu/libm-test-ulps: Update.
	* sysdeps/x86_64/fpu/libm-test-ulps: Likewise.

Comments

Andreas Jaeger March 25, 2014, 12:22 p.m. UTC | #1
On 03/24/2014 05:10 PM, Joseph S. Myers wrote:
> gen-auto-libm-tests presently allows but does not require underflow
> exceptions for results with magnitude in the range (greatest
> subnormal, least normal].
> 
> In some cases, the magnitude of the exact result is very slightly
> above the least normal, but rounding in the implementation results in
> it effectively computing an infinite-precision result that is slightly
> below the least normal, so raising an underflow exception.  This is in
> accordance with the documented accuracy goals, but results in
> testsuite failures.
> 
> This patch changes the logic to allow underflows when the mathematical
> result is up to 0.5ulp above the least normal (so in any case where
> the round-to-nearest result is the least normal).  Ideally underflows
> in all these cases would be accepted only when an underflow with the
> actual result is consistent with the rounding mode (in FE_TOWARDZERO
> mode, a return value of the least normal implies that the
> infinite-precision result did not underflow so there should be no
> underflow exception, for example), so as to match the documented goals
> more precisely - whereas at present the tests for exceptions are
> completely independent of the tests of the returned values.  (The same
> applies to overflow exceptions as well - they too should be checked
> for consistency with the result, as in FE_TOWARDZERO mode a result
> 1ulp below the largest finite value should be inconsistent with an
> overflow exception and cause a failure with overflow rather than
> simply being considered a 1ulp error when overflow is expected.)  But
> the present patch at least deals with the cases causing spurious
> failures so that (a) certain existing tests no longer need to be
> marked as having spurious exceptions (such markings in
> auto-libm-test-in end up applying to more cases than just those they
> are needed for) and (b) log1p can be tested in all rounding modes
> without introducing more such failures.  This patch duly moves tests
> of log1p to ALL_RM_TEST.
> 
> Tested x86_64 and x86 and ulps updated accordingly.

Thanks!

Andreas

Patch
diff mbox

diff --git a/math/auto-libm-test-in b/math/auto-libm-test-in
index 5975631..7c80192 100644
--- a/math/auto-libm-test-in
+++ b/math/auto-libm-test-in
@@ -61,9 +61,8 @@  asin -0x0.ffffffffffffp0
 asin 0x0.ffffffffffffffffp0
 asin -0x0.ffffffffffffffffp0
 # Bug 16351: underflow exception may be missing.
-# Bug 16599: spurious underflow may occur.
 asin min missing-underflow
-asin -min missing-underflow spurious-underflow:dbl-64:x86
+asin -min missing-underflow
 asin min_subnorm missing-underflow
 asin -min_subnorm missing-underflow
 
@@ -175,9 +174,8 @@  atanh 0x1p-5
 atanh 0x1p-10
 atanh 0x1.2345p-20
 # Bug 16352: underflow exception may be missing.
-# Bug 16357: spurious underflow may occur.
-atanh min missing-underflow spurious-underflow:ldbl-96-intel:x86
-atanh -min missing-underflow spurious-underflow:ldbl-96-intel:x86
+atanh min missing-underflow
+atanh -min missing-underflow
 atanh min_subnorm missing-underflow
 atanh -min_subnorm missing-underflow
 
diff --git a/math/gen-auto-libm-tests.c b/math/gen-auto-libm-tests.c
index 61097e4..132cca2 100644
--- a/math/gen-auto-libm-tests.c
+++ b/math/gen-auto-libm-tests.c
@@ -171,6 +171,8 @@  typedef struct
   int min_exp;
   /* The largest normal value.  */
   mpfr_t max;
+  /* The value 0.5ulp above the least positive normal value.  */
+  mpfr_t min_plus_half;
   /* The least positive normal value, 2^(MIN_EXP-1).  */
   mpfr_t min;
   /* The greatest positive subnormal value.  */
@@ -183,13 +185,13 @@  typedef struct
    enumeration.  */
 static fp_format_desc fp_formats[fp_num_formats] =
   {
-    { "flt-32", "f", NULL, 24, 128, -125, {}, {}, {}, {} },
-    { "dbl-64", "", NULL, 53, 1024, -1021, {}, {}, {}, {} },
-    { "ldbl-96-intel", "L", NULL, 64, 16384, -16381, {}, {}, {}, {} },
-    { "ldbl-96-m68k", "L", NULL, 64, 16384, -16382, {}, {}, {}, {} },
-    { "ldbl-128", "L", NULL, 113, 16384, -16381, {}, {}, {}, {} },
+    { "flt-32", "f", NULL, 24, 128, -125, {}, {}, {}, {}, {} },
+    { "dbl-64", "", NULL, 53, 1024, -1021, {}, {}, {}, {}, {} },
+    { "ldbl-96-intel", "L", NULL, 64, 16384, -16381, {}, {}, {}, {}, {} },
+    { "ldbl-96-m68k", "L", NULL, 64, 16384, -16382, {}, {}, {}, {}, {} },
+    { "ldbl-128", "L", NULL, 113, 16384, -16381, {}, {}, {}, {}, {} },
     { "ldbl-128ibm", "L", "0x1.fffffffffffff7ffffffffffff8p+1023",
-      106, 1024, -968, {}, {}, {}, {} },
+      106, 1024, -968, {}, {}, {}, {}, {} },
   };
 
 /* The supported rounding modes.  */
@@ -702,6 +704,10 @@  init_fp_formats ()
       assert_exact (mpfr_set_ui_2exp (fp_formats[f].min, 1,
 				      fp_formats[f].min_exp - 1,
 				      MPFR_RNDN));
+      mpfr_init2 (fp_formats[f].min_plus_half, fp_formats[f].mant_dig + 1);
+      assert_exact (mpfr_set (fp_formats[f].min_plus_half,
+			      fp_formats[f].min, MPFR_RNDN));
+      mpfr_nextabove (fp_formats[f].min_plus_half);
       mpfr_init2 (fp_formats[f].subnorm_max, fp_formats[f].mant_dig);
       assert_exact (mpfr_set (fp_formats[f].subnorm_max, fp_formats[f].min,
 			      MPFR_RNDN));
@@ -1841,8 +1847,9 @@  output_for_one_input_case (FILE *fp, const char *filename, test_function *tf,
 				    <= 0));
 			  may_underflow
 			    |= (!mpfr_zero_p (generic_outputs[i].value.f)
-				&& mpfr_cmpabs (generic_outputs[i].value.f,
-						fp_formats[f].min) <= 0);
+				&& (mpfr_cmpabs (generic_outputs[i].value.f,
+						 fp_formats[f].min_plus_half)
+				    <= 0));
 			}
 		      /* If the result is an exact zero, the sign may
 			 depend on the rounding mode, so recompute it
diff --git a/math/libm-test.inc b/math/libm-test.inc
index a8ebecd..9c4c2bf 100644
--- a/math/libm-test.inc
+++ b/math/libm-test.inc
@@ -7823,9 +7823,7 @@  static const struct test_f_f_data log1p_test_data[] =
 static void
 log1p_test (void)
 {
-  START (log1p, 0);
-  RUN_TEST_LOOP_f_f (log1p, log1p_test_data, );
-  END;
+  ALL_RM_TEST (log1p, 0, log1p_test_data, RUN_TEST_LOOP_f_f, END);
 }
 
 
diff --git a/sysdeps/i386/fpu/libm-test-ulps b/sysdeps/i386/fpu/libm-test-ulps
index 1885be7..a82de19 100644
--- a/sysdeps/i386/fpu/libm-test-ulps
+++ b/sysdeps/i386/fpu/libm-test-ulps
@@ -1288,6 +1288,30 @@  Function: "log1p":
 ildouble: 1
 ldouble: 1
 
+Function: "log1p_downward":
+double: 1
+float: 1
+idouble: 1
+ifloat: 1
+ildouble: 2
+ldouble: 2
+
+Function: "log1p_towardzero":
+double: 1
+float: 1
+idouble: 1
+ifloat: 1
+ildouble: 2
+ldouble: 2
+
+Function: "log1p_upward":
+double: 1
+float: 1
+idouble: 1
+ifloat: 1
+ildouble: 1
+ldouble: 1
+
 Function: "log_downward":
 ildouble: 1
 ldouble: 1
diff --git a/sysdeps/x86_64/fpu/libm-test-ulps b/sysdeps/x86_64/fpu/libm-test-ulps
index 670f2da..e2fa9b2 100644
--- a/sysdeps/x86_64/fpu/libm-test-ulps
+++ b/sysdeps/x86_64/fpu/libm-test-ulps
@@ -1362,6 +1362,30 @@  ifloat: 1
 ildouble: 1
 ldouble: 1
 
+Function: "log1p_downward":
+double: 1
+float: 1
+idouble: 1
+ifloat: 1
+ildouble: 2
+ldouble: 2
+
+Function: "log1p_towardzero":
+double: 1
+float: 1
+idouble: 1
+ifloat: 1
+ildouble: 2
+ldouble: 2
+
+Function: "log1p_upward":
+double: 1
+float: 1
+idouble: 1
+ifloat: 1
+ildouble: 1
+ldouble: 1
+
 Function: "log_downward":
 float: 1
 ifloat: 1