about summary refs log tree commit diff
path: root/math
Commit message (Collapse)AuthorAgeFilesLines
* Support for type-generic libm function implementations libmPaul E. Murphy2016-08-1716-313/+71
| | | | | | | | | | | | | | | | | | | | | | | This defines a new classes of libm objects. The <func>_template.c file which is used in conjunction with the new makefile hooks to derive variants for each type supported by the target machine. The headers math-type-macros-TYPE.h are used to supply macros to a common implementation of a function in a file named FUNC_template.c and glued togethor via a generated file matching existing naming in the build directory. This has the properties of preserving the existing override mechanism and not requiring any arcane build system twiddling. Likewise, it enables machines to override these files without any additional work. I have verified the built objects for ppc64, x86_64, alpha, arm, and m68k do not change in any meaningful way with these changes using the Fedora cross toolchains. I have verified the x86_64 and ppc64 changes still run.
* Add fesetexcept.Joseph Myers2016-08-166-3/+238
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | TS 18661-1 defines an fesetexcept function for setting floating-point exception flags without the side-effect of causing enabled traps to be taken. This patch series implements this function for glibc. The present patch adds the fallback stub implementation, x86 and x86_64 implementations, documentation, tests and ABI baseline updates. The remaining patches, some of them untested, add implementations for other architectures. The implementations generally follow those of the fesetexceptflag function. As for fesetexceptflag, the approach taken for architectures where setting flags causes enabled traps to be taken is to set the flags (and potentially cause traps) rather than refusing to set the flags and returning an error. Since ISO C and TS 18661 provide no way to enable traps, this is formally in accordance with the standards. The NEWS entry should be considered a placeholder, since this patch series is intended to be followed by further such series adding other TS 18661-1 features, so that the NEWS entry would end up looking more like * New <fenv.h> features from TS 18661-1:2014 are added to libm: the fesetexcept, fetestexceptflag, fegetmode and fesetmode functions, the femode_t type and the FE_DFL_MODE macro. with hopefully more such entries for other features, rather than having an entry for a single function in the end. I believe we have consensus for adding TS 18661-1 interfaces as per <https://sourceware.org/ml/libc-alpha/2016-06/msg00421.html>. Tested for x86_64, x86, mips64 (hard float, and soft float to test the fallback version), arm (hard float) and powerpc (hard float, soft float and e500). * math/fesetexcept.c: New file. * sysdeps/i386/fpu/fesetexcept.c: Likewise. * sysdeps/x86_64/fpu/fesetexcept.c: Likewise. * math/fenv.h: Define __GLIBC_INTERNAL_STARTING_HEADER_IMPLEMENTATION and include <bits/libc-header-start.h> instead of including <features.h>. [__GLIBC_USE (IEC_60559_BFP_EXT)] (fesetexcept): New function declaration. * manual/arith.texi (fesetexcept): Document function. * math/Versions (fesetexcept): New libm symbol at version GLIBC_2.25. * math/Makefile (libm-support): Add fesetexcept. (tests): Add test-fesetexcept and test-fesetexcept-traps. * math/test-fesetexcept.c: New file. * math/test-fesetexcept-traps.c: Likewise. * sysdeps/nacl/libm.abilist: Update. * sysdeps/unix/sysv/linux/aarch64/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/alpha/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/arm/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/hppa/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/i386/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/ia64/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/m68k/coldfire/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/m68k/m680x0/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/microblaze/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/mips/mips32/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/mips/mips64/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/nios2/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/powerpc/powerpc32/fpu/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/powerpc/powerpc32/nofpu/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/powerpc/powerpc64/libm-le.abilist: Likewise. * sysdeps/unix/sysv/linux/powerpc/powerpc64/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/s390/s390-32/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/s390/s390-64/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/sh/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/sparc/sparc32/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/sparc/sparc64/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/tile/tilegx/tilegx32/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/tile/tilegx/tilegx64/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/tile/tilepro/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/x86_64/64/libm.abilist: Likewise. * sysdeps/unix/sysv/linux/x86_64/x32/libm.abilist: Likewise.
* Fix test-fexcept when "inexact" implicitly raised.Joseph Myers2016-08-121-2/+31
| | | | | | | | | | | | | | | | | | ISO C allows feraiseexcept to raise "inexact", in addition to the requested exceptions, when requested to raise "overflow" or "underflow". Testing on ARM and PowerPC e500 (where glibc's feraiseexcept has this property) showed that the new test-fexcept test failed to allow for this; this patch fixes it, by wrapping feraiseexcept to clear FE_INEXACT if implicitly raised and not raised before the call. (It would also be possible to do this with fesetexcept, which always affects exactly the requested flags, but this patch avoids making this fix depend on the fesetexcept changes.) Tested for x86_64, x86, arm and e500. * math/test-fexcept.c (feraiseexcept_exact): New function. (test_set): Call feraiseexcept_exact instead of feraiseexcept. (test_except): Likewise.
* Add tests for fegetexceptflag, fesetexceptflag.Joseph Myers2016-08-103-1/+256
| | | | | | | | | | | | | | | | | | | | | | | | | | | | I noticed that there was no meaningful test coverage for fegetexceptflag and fesetexceptflag (one test ensures that calls to them compile and link, but nothing to verify they work correctly). This patch adds tests for these functions. fesetexceptflag is meant to set the relevant exception flag bits to the saved state without causing enabled traps to be taken. On some architectures, it is not possible to set exception flag bits without causing enabled traps to occur. Such architectures need to define EXCEPTION_SET_FORCES_TRAP to 1 in their math-tests.h, as is done in this patch for powerpc. x86 avoids needing to define this because the traps resulting from setting exception bits don't occur until the next floating-point operation or fwait instruction. Tested for x86_64, x86 and powerpc. Note that test-fexcept fails for powerpc because of a pre-existing bug in fesetexceptflag for powerpc, which I'll fix separately. * math/test-fexcept-traps.c: New file. * math/test-fexcept.c: Likewise. * math/Makefile (tests): Add test-fexcept and test-fexcept-traps. * sysdeps/generic/math-tests.h (EXCEPTION_SET_FORCES_TRAP): New macro. * sysdeps/powerpc/math-tests.h [!__NO_FPRS__] (EXCEPTION_SET_FORCES_TRAP): Likewise.
* sparc32/sparcv9: add a VIS3 version of fdimAurelien Jarno2016-08-052-1/+5
| | | | | | | | | | | | | | | | | | | sparc32 passes floating point values in the integer registers. VIS3 instructions gives access to the movwtos instruction to directly transfer a value from an integer register to a floating point register. Therefore it makes sense to provide a VIS3 version consisting in the generic version compiled with -mvis3. Changelog: * math/s_fdim.c: Avoid alias renamed. * math/s_fdimf.c: Likewise. * sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/Makefile [$(subdir) = math && $(have-as-vis3) = yes] (libm-sysdep_routines): Add s_fdimf-vis3, s_fdim-vis3. (CFLAGS-s_fdimf-vis3.c): New. Set to -Wa,-Av9d -mvis3. (CFLAGS-s_fdim-vis3.c): Likewise. sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim-vis3.c: New file. sysdeps/sparc/sparc32/sparcv9/fpu/multiarch/s_fdim.c: Likewise.
* Improve gen-libm-test.pl LIT() applicationPaul E. Murphy2016-08-052-56/+71
| | | | | | | | | | | | | | | | | | When bootstrapping float128, this exposed a number of areas where the L suffix is incorrectly applied to simple expressions when it should be applied to each constant in the expression. In order to stave off more macros in libm-test.inc, apply_lit is made slightly more intelligent. It will now split expressions based on space characters, and attempt to apply LIT() to each token. Having done this, there are numerous spacing issues within libm-test.inc which have been fixed. The above is problematic when the L real suffix is not the most expressive modifier, and the compiler complains (i.e ppc64) or silently truncates a value (i.e ppc64).
* Fix math.h comment about bits/mathdef.h.Joseph Myers2016-08-051-34/+21
| | | | | | | | | | | | | | | | | | | | | | | | | | | | math.h has a comment about definitions from <bits/mathdef.h>. This comment is in the wrong place in math.h, far below the inclusion of <bits/mathdef.h>. It was originally above the inclusion, but the inclusion was moved by 1998-11-05 Ulrich Drepper <drepper@cygnus.com> * math/math.h: Unconditionally include bits/mathdef.h. Declare long double functions only if __NO_LONG_DOUBLE_MATH is not defined. [...] without moving the comment. Furthermore, the comment refers incorrectly to FLT_EVAL_METHOD and DECIMAL_DIG, which are actually <float.h> macros, and INFINITY, which is in <bits/inf.h>. This patch moves the comment back above the include it refers to and removes the description of macros not defined by the header. Tested for x86_64 and x86 (testsuite, and that installed stripped shared libraries are unchanged by the patch). * math/math.h: Move comment about <bits/mathdef.h> definitions above inclusion of <bits/mathdef.h>. Do not mention FLT_EVAL_METHOD, INFINITY or DECIMAL_DIG in that comment.
* Do not call __nan in scalb functions.Joseph Myers2016-08-043-15/+3
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | When libm functions return a NaN: if it is for NaN input, it should be computed from that input (e.g. adding it to itself), so that payloads are propagated and signaling NaNs quieted, while if it is for non-NaN input, it should be produced by a computation such as (x - x) / (x - x), which raises "invalid" at the same time as producing an appropriate NaN, so avoiding any need for a call to feraiseexcept. Various libm functions, however, call __nan ("") (or __nanf or __nanl) to determine the NaN to return, together with using feraiseexcept (FE_INVALID) to raise the exception. sysdeps/generic/math_private.h has an optimization for those functions with constant "" argument so this doesn't actually involve a call to the __nan function, but it is still not the preferred approach for producing NaNs. (The optimized code also always uses the NAN macro, i.e. produces a default NaN for float converted to whatever the target type is, and on some architectures that may not be the same as the preferred default NaN for double or long double.) This patch fixes the scalb functions to use the conventional method of generating NaNs and raising "invalid" with an appropriate computation. (Most instances of this issue are in the complex functions, where it can more readily be fixed once they have been made type-generic and so only a third as many places need fixing. Some of the complex functions use __nan ("") + __nan (""), where the addition serves no purpose whatsoever.) Tested for x86_64 and x86. * math/e_scalb.c: Do not include <fenv.h>. (invalid_fn): Do calculation resulting in NaN instead of raising FE_INVALID and returning a NaN explicitly. * math/e_scalbf.c: Do not include <fenv.h>. (invalid_fn): Do calculation resulting in NaN instead of raising FE_INVALID and returning a NaN explicitly. * math/e_scalbl.c: Do not include <fenv.h>. (invalid_fn): Do calculation resulting in NaN instead of raising FE_INVALID and returning a NaN explicitly.
* Also handle __STDC_WANT_IEC_60559_BFP_EXT__ in <tgmath.h>.Joseph Myers2016-08-031-1/+1
| | | | | | | | | | | | | | | | My __STDC_WANT_IEC_60559_BFP_EXT__ patch omitted to update the conditions on the nextup and nextdown type-generic macros in <tgmath.h>. This patch updates those conditions accordingly. (As glibc doesn't currently have an exp10 type-generic macro, no such changes are needed relating to __STDC_WANT_IEC_60559_FUNCS_EXT__; adding such a type-generic macro would be a new feature.) Tested for x86_64 and x86 (testsuite, and that installed stripped shared libraries are unchanged by the patch). Committed. * math/tgmath.h (nextdown): Define if [__GLIBC_USE (IEC_60559_BFP_EXT)], not if [__USE_GNU]. (nextup): Likewise.
* Support __STDC_WANT_IEC_60559_FUNCS_EXT__ feature test macro.Joseph Myers2016-08-031-2/+4
| | | | | | | | | | | | | | | | | | | | | | | This patch implements support for the __STDC_WANT_IEC_60559_FUNCS_EXT__ feature test macro, following the __GLIBC_USE approach used for other ISO C feature test macros. Currently this only affects the exp10 functions (which glibc has had for a long time). Tested for x86_64 and x86 (testsuite, and that installed stripped shared libraries are unchanged by the patch). * bits/libc-header-start.h (__GLIBC_USE_IEC_60559_FUNCS_EXT): New macro. * include/features.h (__STDC_WANT_IEC_60559_FUNCS_EXT__): Document. * manual/creature.texi (__STDC_WANT_IEC_60559_FUNCS_EXT__): Document macro. * manual/math.texi (exp10): Document as ISO from TS 18661-4:2015. (exp10f): Likewise. (exp10l): Likewise. * math/bits/mathcalls.h (exp10): Declare if [__GLIBC_USE (IEC_60559_FUNCS_EXT)], not [__USE_GNU].
* Support __STDC_WANT_IEC_60559_BFP_EXT__ feature test macro.Joseph Myers2016-08-032-5/+6
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | This patch implements support for the __STDC_WANT_IEC_60559_BFP_EXT__ feature test macro from ISO/IEC 18661-1:2014, following the __GLIBC_USE approach now used for __STDC_WANT_LIB_EXT2__. For this macro, the relevant consideration is whether it is defined or undefined when an affected header is included (not what its value is if defined, and not whether it's defined or undefined when any other unaffected system header is included). Currently this macro only affects the issignaling macro and the nextup and nextdown functions (so they can be enabled by defining this macro, not just by defining _GNU_SOURCE as previously). Any further features from this TS added in future would also be conditioned on this macro. Tested for x86_64 and x86 (testsuite, and that installed stripped shared libraries are unchanged by the patch). * bits/libc-header-start.h (__GLIBC_USE_IEC_60559_BFP_EXT): New macro. * include/features.h (__STDC_WANT_IEC_60559_BFP_EXT__): Document. * manual/arith.texi (issignaling): Document as ISO from TS 18661-1:2014. (nextup): Likewise. (nextupf): Likewise. (nextupl): Likewise. (nextdown): Likewise. (nextdownf): Likewise. (nextdownl): Likewise. * manual/creature.texi (__STDC_WANT_IEC_60559_BFP_EXT__): Document macro. * math/math.h: Define __GLIBC_INTERNAL_STARTING_HEADER_IMPLEMENTATION and include <bits/libc-header-start.h> instead of including <features.h>. (issignaling): Define if [__GLIBC_USE (IEC_60559_BFP_EXT)], not [__USE_GNU]. * math/bits/mathcalls.h (nextdown): Declare if [__GLIBC_USE (IEC_60559_BFP_EXT)], not [__USE_GNU]. (nextup): Likewise. (__issignaling): Likewise.
* Unify drift between _Complex function type variantsPaul E. Murphy2016-08-034-10/+4
| | | | | | While trying to convert the _Complex function wrappers into a single generic implementation, a few minor variations between identical versions emerged.
* Refactor part of math MakefilePaul E. Murphy2016-08-032-56/+75
| | | | | | | | | | | | | | | | | | | | | | | In order to support more types, the Makefile needs a few bits shuffled. F is explictly used as a placeholder to substitute for the appropriate type suffix. This removes the need to demangle _r suffixed objects. The variable libm-compat-calls is added to house any objects which are only built to provide compat symbols within libm. That is, no newly added type should ever attempt building these. Note, k_standard* files have been added there. By consensus they are deprecated; in practice, we haven't gotten there yet. New types would be added as noted in the comments preceding type-TYPE-{suffix,routines,yes} variables. However, some manual additions will still need to be done to add appropriate flags when building the various variants of libm-test.c for a new type. Likewise, test-ildoubl is renamed test-ildouble for consistency's sake.
* x86_64: Call finite scalar versions in vectorized log, pow, exp (bz #20033).Andrew Senkevich2016-08-021-4/+8
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Vector math functions require -ffast-math which sets -ffinite-math-only, so it is needed to call finite scalar versions (which are called from vector functions in some cases). Since finite version of pow() returns qNaN instead of 1.0 for several inputs, those inputs are excluded for tests of vector math functions. [BZ #20033] * sysdeps/x86_64/fpu/multiarch/svml_d_exp2_core_sse4.S: Call finite version. * sysdeps/x86_64/fpu/multiarch/svml_d_exp4_core_avx2.S: Likewise. * sysdeps/x86_64/fpu/multiarch/svml_d_exp8_core_avx512.S: Likewise. * sysdeps/x86_64/fpu/multiarch/svml_d_log2_core_sse4.S: Likewise. * sysdeps/x86_64/fpu/multiarch/svml_d_log4_core_avx2.S: Likewise. * sysdeps/x86_64/fpu/multiarch/svml_d_log8_core_avx512.S: Likewise. * sysdeps/x86_64/fpu/multiarch/svml_d_pow2_core_sse4.S: Likewise. * sysdeps/x86_64/fpu/multiarch/svml_d_pow4_core_avx2.S: Likewise. * sysdeps/x86_64/fpu/multiarch/svml_d_pow8_core_avx512.S: Likewise. * sysdeps/x86_64/fpu/multiarch/svml_s_expf16_core_avx512.S: Likewise. * sysdeps/x86_64/fpu/multiarch/svml_s_expf4_core_sse4.S: Likewise. * sysdeps/x86_64/fpu/multiarch/svml_s_expf8_core_avx2.S: Likewise. * sysdeps/x86_64/fpu/multiarch/svml_s_logf16_core_avx512.S: Likewise. * sysdeps/x86_64/fpu/multiarch/svml_s_logf4_core_sse4.S: Likewise. * sysdeps/x86_64/fpu/multiarch/svml_s_logf8_core_avx2.S: Likewise. * sysdeps/x86_64/fpu/multiarch/svml_s_powf16_core_avx512.S: Likewise. * sysdeps/x86_64/fpu/multiarch/svml_s_powf4_core_sse4.S: Likewise. * sysdeps/x86_64/fpu/multiarch/svml_s_powf8_core_avx2.S: Likewise. * sysdeps/x86_64/fpu/svml_d_exp2_core.S: Likewise. * sysdeps/x86_64/fpu/svml_d_log2_core.S: Likewise. * sysdeps/x86_64/fpu/svml_d_pow2_core.S: Likewise. * sysdeps/x86_64/fpu/svml_s_expf4_core.S: Likewise. * sysdeps/x86_64/fpu/svml_s_logf4_core.S: Likewise. * sysdeps/x86_64/fpu/svml_s_powf4_core.S: Likewise. * math/libm-test.inc (pow_test_data): Exclude tests for qNaN in power zero.
* Fix cos computation for multiple precision fallback (bz #20357)Siddhesh Poyarekar2016-07-182-0/+210
| | | | | | | | | | | | | | | | | | | | During the sincos consolidation I made two mistakes, one was a logical error due to which cos(0x1.8475e5afd4481p+0) returned sin(0x1.8475e5afd4481p+0) instead. The second issue was an error in negating inputs for the correct quadrants for sine. I could not find a suitable test case for this despite running a program to search for such an input for a couple of hours. Following patch fixes both issues. Tested on x86_64. Thanks to Matt Clay for identifying the issue. [BZ #20357] * sysdeps/ieee754/dbl-64/s_sin.c (sloww): Fix up condition to call __mpsin/__mpcos and to negate values. * math/auto-libm-test-in: Add test. * math/auto-libm-test-out: Regenerate.
* Avoid "inexact" exceptions in i386/x86_64 trunc functions (bug 15479).Joseph Myers2016-06-271-76/+75
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | As discussed in <https://sourceware.org/ml/libc-alpha/2016-05/msg00577.html>, TS 18661-1 disallows ceil, floor, round and trunc functions from raising the "inexact" exception, in accordance with general IEEE 754 semantics for when that exception is raised. Fixing this for x87 floating point is more complicated than for the other versions of these functions, because they use the frndint instruction that raises "inexact" and this can only be avoided by saving and restoring the whole floating-point environment. As I noted in <https://sourceware.org/ml/libc-alpha/2016-06/msg00128.html>, I have now implemented a GCC option -fno-fp-int-builtin-inexact for GCC 7, such that GCC will inline these functions on x86, without caring about "inexact", when the default -ffp-int-builtin-inexact is in effect. This allows users to get optimized code depending on the options they pass to the compiler, while making the out-of-line functions follow TS 18661-1 semantics and avoid "inexact". This patch duly fixes the out-of-line trunc function implementations to avoid "inexact", in the same way as the nearbyint implementations. I do not know how the performance of implementations such as these based on saving the environment and changing the rounding mode temporarily compares to that of the C versions or SSE 4.1 versions (of course, for 32-bit x86 SSE implementations still need to get the return value in an x87 register); it's entirely possible other implementations could be faster in some cases. Tested for x86_64 and x86. [BZ #15479] * sysdeps/i386/fpu/s_trunc.S (__trunc): Save and restore floating-point environment rather than just control word. * sysdeps/i386/fpu/s_truncf.S (__truncf): Likewise. * sysdeps/i386/fpu/s_truncl.S (__truncl): Save and restore floating-point environment, with "invalid" exceptions merged in, rather than just control word. * sysdeps/x86_64/fpu/s_truncl.S (__truncl): Likewise. * math/libm-test.inc (trunc_test_data): Do not allow spurious "inexact" exceptions.
* Avoid "inexact" exceptions in i386/x86_64 floor functions (bug 15479).Joseph Myers2016-06-271-71/+70
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | As discussed in <https://sourceware.org/ml/libc-alpha/2016-05/msg00577.html>, TS 18661-1 disallows ceil, floor, round and trunc functions from raising the "inexact" exception, in accordance with general IEEE 754 semantics for when that exception is raised. Fixing this for x87 floating point is more complicated than for the other versions of these functions, because they use the frndint instruction that raises "inexact" and this can only be avoided by saving and restoring the whole floating-point environment. As I noted in <https://sourceware.org/ml/libc-alpha/2016-06/msg00128.html>, I have now implemented a GCC option -fno-fp-int-builtin-inexact for GCC 7, such that GCC will inline these functions on x86, without caring about "inexact", when the default -ffp-int-builtin-inexact is in effect. This allows users to get optimized code depending on the options they pass to the compiler, while making the out-of-line functions follow TS 18661-1 semantics and avoid "inexact". This patch duly fixes the out-of-line floor function implementations to avoid "inexact", in the same way as the nearbyint implementations. I do not know how the performance of implementations such as these based on saving the environment and changing the rounding mode temporarily compares to that of the C versions or SSE 4.1 versions (of course, for 32-bit x86 SSE implementations still need to get the return value in an x87 register); it's entirely possible other implementations could be faster in some cases. Tested for x86_64 and x86. [BZ #15479] * sysdeps/i386/fpu/s_floor.S (__floor): Save and restore floating-point environment rather than just control word. * sysdeps/i386/fpu/s_floorf.S (__floorf): Likewise. * sysdeps/i386/fpu/s_floorl.S (__floorl): Save and restore floating-point environment, with "invalid" exceptions merged in, rather than just control word. * sysdeps/x86_64/fpu/s_floorl.S (__floorl): Likewise. * math/libm-test.inc (floor_test_data): Do not allow spurious "inexact" exceptions.
* Avoid "inexact" exceptions in i386/x86_64 ceil functions (bug 15479).Joseph Myers2016-06-271-72/+71
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | As discussed in <https://sourceware.org/ml/libc-alpha/2016-05/msg00577.html>, TS 18661-1 disallows ceil, floor, round and trunc functions from raising the "inexact" exception, in accordance with general IEEE 754 semantics for when that exception is raised. Fixing this for x87 floating point is more complicated than for the other versions of these functions, because they use the frndint instruction that raises "inexact" and this can only be avoided by saving and restoring the whole floating-point environment. As I noted in <https://sourceware.org/ml/libc-alpha/2016-06/msg00128.html>, I have now implemented a GCC option -fno-fp-int-builtin-inexact for GCC 7, such that GCC will inline these functions on x86, without caring about "inexact", when the default -ffp-int-builtin-inexact is in effect. This allows users to get optimized code depending on the options they pass to the compiler, while making the out-of-line functions follow TS 18661-1 semantics and avoid "inexact". This patch duly fixes the out-of-line ceil function implementations to avoid "inexact", in the same way as the nearbyint implementations. I do not know how the performance of implementations such as these based on saving the environment and changing the rounding mode temporarily compares to that of the C versions or SSE 4.1 versions (of course, for 32-bit x86 SSE implementations still need to get the return value in an x87 register); it's entirely possible other implementations could be faster in some cases. Tested for x86_64 and x86. [BZ #15479] * sysdeps/i386/fpu/s_ceil.S (__ceil): Save and restore floating-point environment rather than just control word. * sysdeps/i386/fpu/s_ceilf.S (__ceilf): Likewise. * sysdeps/i386/fpu/s_ceill.S (__ceill): Save and restore floating-point environment, with "invalid" exceptions merged in, rather than just control word. * sysdeps/x86_64/fpu/s_ceill.S (__ceill): Likewise. * math/libm-test.inc (ceil_test_data): Do not allow spurious "inexact" exceptions.
* Fix i386/x86_64 scalbl with sNaN input (bug 20296).Joseph Myers2016-06-231-0/+36
| | | | | | | | | | | | | | | | | The x86_64 and i386 versions of scalbl return sNaN for some cases of sNaN input and are missing "invalid" exceptions for other cases. This results from overly complicated code that either returns a NaN input, or discards both inputs when one is NaN and loads a NaN from memory. This patch fixes this by simplifying the code to add the arguments when either one is NaN. Tested for x86_64 and x86. [BZ #20296] * sysdeps/i386/fpu/e_scalbl.S (__ieee754_scalbl): Add arguments when either argument is a NaN. * sysdeps/x86_64/fpu/e_scalbl.S (__ieee754_scalbl): Likewise. * math/libm-test.inc (scalb_test_data): Add sNaN tests.
* Add more sNaN tests (most remaining real functions).Joseph Myers2016-06-232-0/+254
| | | | | | | | | | | | | | | | | | | | | | | | This patch adds tests of sNaN inputs to more functions to libm-test.inc. This covers the remaining real functions except for scalb, where there's a bug to fix, and hypot pow fmin fmax, where there are cases where a qNaN input does not result in a qNaN output and so sNaN support according to TS 18661-1 is more of a new feature. Tested for x86_64 and x86. * math/libm-test.inc (snan_value_ld): New macro. (isgreater_test_data): Add sNaN tests. (isgreaterequal_test_data): Likewise. (isless_test_data): Likewise. (islessequal_test_data): Likewise. (islessgreater_test_data): Likewise. (isunordered_test_data): Likewise. (nextafter_test_data): Likewise. (nexttoward_test_data): Likewise. (remainder_test_data): Likewise. (remquo_test_data): Likewise. (significand_test_data): Likewise. * math/gen-libm-test.pl (%beautify): Add snan_value_ld.
* Add nextup and nextdown math functionsRajalakshmi Srinivasaraghavan2016-06-169-7/+253
| | | | | | | | | | | TS 18661 adds nextup and nextdown functions alongside nextafter to provide support for float128 equivalent to it. This patch adds nextupl, nextup, nextupf, nextdownl, nextdown and nextdownf to libm before float128 support. The nextup functions return the next representable value in the direction of positive infinity and the nextdown functions return the next representable value in the direction of negative infinity. These are currently enabled as GNU extensions.
* Fix i386 fdim double rounding (bug 20255).Joseph Myers2016-06-141-0/+3
| | | | | | | | | | | | | | | | | | | fdim suffers from double rounding on i386 because subtracting two double values can produce an inexact long double value exactly half way between two double values. This patch fixes this by creating an i386-specific version of fdim - C, based on the generic version, unlike the previous .S version - which sets the x87 precision control to double precision for the subtraction and then restores it afterwards. As noted in the comment added, there are no issues of double rounding for subnormals (a case that setting precision control does not address) because subtraction cannot produce an inexact result in the subnormal range. Tested for x86_64 and x86. [BZ #20255] * sysdeps/i386/fpu/s_fdim.c: New file. Based on math/s_fdim.c. * math/libm-test.inc (fdim_test_data): Add another test.
* Use generic fdim on more architectures (bug 6796, bug 20255, bug 20256).Joseph Myers2016-06-143-4/+37
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Some architectures have their own versions of fdim functions, which are missing errno setting (bug 6796) and may also return sNaN instead of qNaN for sNaN input, in the case of the x86 / x86_64 long double versions (bug 20256). These versions are not actually doing anything that a compiler couldn't generate, just straightforward comparisons / arithmetic (and, in the x86 / x86_64 case, testing for NaNs with fxam, which isn't actually needed once you use an unordered comparison and let the NaNs pass through the same subtraction as non-NaN inputs). This patch removes the x86 / x86_64 / powerpc versions, so that those architectures use the generic C versions, which correctly handle setting errno and deal properly with sNaN inputs. This seems better than dealing with setting errno in lots of .S versions. The i386 versions also return results with excess range and precision, which is not appropriate for a function exactly defined by reference to IEEE operations. For errno setting to work correctly on overflow, it's necessary to remove excess range with math_narrow_eval, which this patch duly does in the float and double versions so that the tests can reliably pass on x86. For float, this avoids any double rounding issues as the long double precision is more than twice that of float. For double, double rounding issues will need to be addressed separately, so this patch does not fully fix bug 20255. Tested for x86_64, x86 and powerpc. [BZ #6796] [BZ #20255] [BZ #20256] * math/s_fdim.c: Include <math_private.h>. (__fdim): Use math_narrow_eval on result. * math/s_fdimf.c: Include <math_private.h>. (__fdimf): Use math_narrow_eval on result. * sysdeps/i386/fpu/s_fdim.S: Remove file. * sysdeps/i386/fpu/s_fdimf.S: Likewise. * sysdeps/i386/fpu/s_fdiml.S: Likewise. * sysdeps/i386/i686/fpu/s_fdim.S: Likewise. * sysdeps/i386/i686/fpu/s_fdimf.S: Likewise. * sysdeps/i386/i686/fpu/s_fdiml.S: Likewise. * sysdeps/powerpc/fpu/s_fdim.c: Likewise. * sysdeps/powerpc/fpu/s_fdimf.c: Likewise. * sysdeps/powerpc/powerpc32/fpu/s_fdim.c: Likewise. * sysdeps/powerpc/powerpc64/fpu/s_fdim.c: Likewise. * sysdeps/x86_64/fpu/s_fdiml.S: Likewise. * math/libm-test.inc (fdim_test_data): Expect errno setting on overflow. Add sNaN tests.
* Simplify generic fdim implementations.Joseph Myers2016-06-143-30/+6
| | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | The generic fdim implementations have unnecessarily complicated code, using fpclassify to determine whether the arguments are NaNs, subtracting NaNs if so and otherwise subtracting the non-NaN arguments if not (x <= y), then using fpclassify on the result to see if it is infinite. This patch simplifies the code. Instead of handling NaNs separately, it suffices to use an unordered comparison with islessequal (x, y) to determine whether to return zero, and otherwise NaNs can go through the same subtraction as non-NaN arguments; no explicit tests for NaN are needed at all. Then, isinf instead of fpclassify can be used to determine whether to set errno (in the normal non-overflow case, only one classification will need to occur, unlike the three in the previous code, of which two occurred even if returning zero, because the result will not be infinite in the normal case). The resulting logic is essentially the same as that in the powerpc version, except that the powerpc version is missing errno setting and uses <= not islessequal, so relying on <https://gcc.gnu.org/bugzilla/show_bug.cgi?id=58684>, the GCC bug that unordered comparison instructions are wrongly used on powerpc for ordered comparisons. The compiled code for fdim and fdimf on x86_64 is less than half the size of the previous code. Tested for x86_64. * math/s_fdim.c (__fdim): Use islessequal and isinf instead of fpclassify. * math/s_fdimf.c (__fdimf): Likewise. * math/s_fdiml.c (__fdiml): Likewise.
* Fix dbl-64 atan2 (sNaN, qNaN) (bug 20252).Joseph Myers2016-06-131-0/+62
| | | | | | | | | | | | | | | | | | The dbl-64 implementation of atan2, passed arguments (sNaN, qNaN), fails to raise the "invalid" exception. This patch fixes it to add both arguments, rather than just adding the second argument to itself, in the case where the second argument is a NaN (which is checked for before checking for the first argument being a NaN). sNaN tests for atan2 are added, along with some qNaN tests I noticed were missing but should have been there by analogy with other tests present. Tested for x86_64 and x86. [BZ #20252] * sysdeps/ieee754/dbl-64/e_atan2.c (__ieee754_atan2): Add both arguments when second argument is a NaN. * math/libm-test.inc (atan2_test_data): Add sNaN tests and more qNaN tests.
* Add more sNaN tests (cimag, conj, copysign, creal, fma, fmod).Joseph Myers2016-06-131-0/+95
| | | | | | | | | | | | | This patch adds tests of sNaN inputs to further libm functions. Tested for x86_64 and x86. * math/libm-test.inc (cimag_test_data): Add sNaN tests. (conj_test_data): Likewise. (copysign_test_data): Likewise. (creal_test_data): Likewise. (fma_test_data): Likewise. (fmod_test_data): Likewise.
* Fix frexp (NaN) (bug 20250).Joseph Myers2016-06-131-0/+2
| | | | | | | | | | | | | | | | | | | | | Various implementations of frexp functions return sNaN for sNaN input. This patch fixes them to add such arguments to themselves so that qNaN is returned. Tested for x86_64, x86, mips64 and powerpc. [BZ #20250] * sysdeps/i386/fpu/s_frexpl.S (__frexpl): Add non-finite input to itself. * sysdeps/ieee754/dbl-64/s_frexp.c (__frexp): Add non-finite or zero input to itself. * sysdeps/ieee754/dbl-64/wordsize-64/s_frexp.c (__frexp): Likewise. * sysdeps/ieee754/flt-32/s_frexpf.c (__frexpf): Likewise. * sysdeps/ieee754/ldbl-128/s_frexpl.c (__frexpl): Likewise. * sysdeps/ieee754/ldbl-128ibm/s_frexpl.c (__frexpl): Likewise. * sysdeps/ieee754/ldbl-96/s_frexpl.c (__frexpl): Likewise. * math/libm-test.inc (frexp_test_data): Add sNaN tests.
* Fix modf (sNaN) (bug 20240).Joseph Myers2016-06-102-0/+8
| | | | | | | | | | | | | | | | | | | | | Various modf implementations return sNaN (both outputs) for sNaN input. In fact they contain code to convert sNaN to qNaN for both outputs, but the way this is done is multiplying by 1.0 (for a wider range of inputs that includes NaNs as well as numbers with exponent large enough to ensure that they are integers), and that multiplication by 1.0 is optimized away by GCC in the absence of -fsignaling-nans, unlike other operations on NaNs used for this purpose that are not no-ops for non-sNaN input. This patch arranges for those files to be built with -fsignaling-nans so that this existing code is effective as intended. Tested for x86_64 and x86. [BZ #20240] * math/Makefile (CFLAGS-s_modf.c): New variable. (CFLAGS-s_modff.c): Likewise. (CFLAGS-s_modfl.c): Likewise. * math/libm-test.inc (modf_test_data): Add sNaN tests.
* Fix i386/x86_64 log2l (sNaN) (bug 20235).Joseph Myers2016-06-091-0/+2
| | | | | | | | | | | | | | The i386/x86_64 versions of log2l return sNaN for sNaN input. This patch fixes them to add NaN inputs to themselves so that qNaN is returned in this case. Tested for x86_64 and x86. [BZ #20235] * sysdeps/i386/fpu/e_log2l.S (__ieee754_log2l): Add NaN input to itself. * sysdeps/x86_64/fpu/e_log2l.S (__ieee754_log2l): Likewise. * math/libm-test.inc (log2_test_data): Add sNaN tests.
* Fix i386/x86_64 log1pl (sNaN) (bug 20229).Joseph Myers2016-06-081-0/+2
| | | | | | | | | | | | | The i386/x86_64 versions of log1pl return sNaN for sNaN input. This patch fixes them to add a NaN input to itself so that qNaN is returned in this case. Tested for x86_64 and x86. [BZ #20229] * sysdeps/i386/fpu/s_log1pl.S (__log1pl): Add NaN input to itself. * sysdeps/x86_64/fpu/s_log1pl.S (__log1pl): Likewise. * math/libm-test.inc (log1p_test_data): Add sNaN tests.
* Fix i386/x86_64 log10l (sNaN) (bug 20228).Joseph Myers2016-06-081-0/+2
| | | | | | | | | | | | | | The i386/x86_64 versions of log10l return sNaN for sNaN input. This patch fixes them to add a NaN input to itself so that qNaN is returned in this case. Tested for x86_64 and x86. [BZ #20228] * sysdeps/i386/fpu/e_log10l.S (__ieee754_log10l): Add NaN input to itself. * sysdeps/x86_64/fpu/e_log10l.S (__ieee754_log10l): Likewise. * math/libm-test.inc (log10_test_data): Add sNaN tests.
* Fix i386/x86_64 logl (sNaN) (bug 20227).Joseph Myers2016-06-081-0/+2
| | | | | | | | | | | | | | | | The i386/x86_64 versions of logl return sNaN for sNaN input. This patch fixes them to add a NaN input to itself so that qNaN is returned in this case. Tested for x86_64 and x86 (including a build for i586 to cover the non-i686 logl version). [BZ #20227] * sysdeps/i386/fpu/e_logl.S (__ieee754_logl): Add NaN input to itself. * sysdeps/i386/i686/fpu/e_logl.S (__ieee754_logl): Likewise. * sysdeps/x86_64/fpu/e_logl.S (__ieee754_logl): Likewise. * math/libm-test.inc (log_test_data): Add sNaN tests.
* Fix i386/x86_64 expl, exp10l, expm1l for sNaN input (bug 20226).Joseph Myers2016-06-081-0/+6
| | | | | | | | | | | | | | | | | The i386 and x86_64 implementations of expl, exp10l and expm1l (code shared between the functions) return sNaN for sNaN input. This patch fixes them to add NaN inputs to themselves so that qNaN is returned in this case. Tested for x86_64 and x86. [BZ #20226] * sysdeps/i386/fpu/e_expl.S (IEEE754_EXPL): Add NaN argument to itself. * sysdeps/x86_64/fpu/e_expl.S (IEEE754_EXPL): Likewise. * math/libm-test.inc (exp_test_data): Add sNaN tests. (exp10_test_data): Likewise. (expm1_test_data): Likewise.
* Fix ldexp, scalbn, scalbln for sNaN input (bug 20225).Joseph Myers2016-06-087-6/+10
| | | | | | | | | | | | | | | | | | | | The wrapper implementations of ldexp / scalbn / scalbln (architecture-independent), and their float / long double variants, return sNaN for sNaN input. This patch fixes them to add relevant arguments to themselves so that qNaN is returned in this case. Tested for x86_64 and x86. [BZ #20225] * math/s_ldexp.c (__ldexp): Add non-finite or zero argument to itself. * math/s_ldexpf.c (__ldexpf): Likewise. * math/s_ldexpl.c (__ldexpl): Likewise. * math/w_scalbln.c (__w_scalbln): Likewise. * math/w_scalblnf.c (__w_scalblnf): Likewise. * math/w_scalblnl.c (__w_scalblnl): Likewise. * math/libm-test.inc (scalbn_test_data): Add sNaN tests. (scalbln_test_data): Likewise.
* Fix i386 cbrtl (sNaN) (bug 20224).Joseph Myers2016-06-081-0/+2
| | | | | | | | | | | | | | | The i386 version of cbrtl returns sNaN (without raising any exceptions) for sNaN input. This patch fixes it to add non-finite arguments to themselves (the code path in question is also reached for zero arguments, for which adding them to themselves is also harmless), so that "invalid" is raised and qNaN returned. Tested for x86_64 and x86. [BZ #20224] * sysdeps/i386/fpu/s_cbrtl.S (__cbrtl): Add non-finite or zero argument to itself. * math/libm-test.inc (cbrt_test_data): Add sNaN tests.
* Generate new format names in auto-libm-test-outPaul E. Murphy2016-06-083-307667/+307655
| | | | | | | | This converts the inclusion macro for each test to use the format specific macro. In addition, the format specifier is removed as it is applied via the LIT() macro which is itself applied when converting the auto inputs and libm-test.inc into libm-test.c.
* Remove type specific information from auto-libm-test-inPaul E. Murphy2016-06-082-2864/+2865
| | | | | | | | | | | | | Apply the following sed regexes to auto-libm-test-in in order: s/flt-32/binary32/ s/dbl-64/binary64/ s/ldbl-96-intel/intel96/ s/ldbl-96-m68k/m68k96/ s/ldbl-128ibm/ibm128/ s/ldbl-128/binary128/ and fixup ldbl-96 comment manually.
* Remove CHOOSE() macro from libm-tests.incPaul E. Murphy2016-06-0820-39/+41
| | | | | | | | | | Use gen-libm-test.pl to generate a list of macros mapping to libm-test-ulps.h as this simplifies adding new types without having to modify a growing number of static headers each time a type is added. This also removes the final usage of the TEST_(DOUBLE|FLOAT|LDOUBLE) macros. Thus, they too are removed.
* Apply LIT(x) to floating point literals in libm-test.cPaul E. Murphy2016-06-082-95/+133
| | | | | | | | | | | | | | | | | | With the exception of the second argument of nexttoward, any suffixes should be stripped from the test input, and the macro LIT(x) should be applied to use the correct suffix for the type being tested. This adds a new argument type "j" to gen-test-libm.pl to signify an argument to a test input which does not require fixup. The test cases of nexttoward have been updated to use this new feature. This applies post-processing to all of the test inputs through gen-libm-test.pl to strip literal suffixes and apply the LIT(x) macro, with one exception stated above. This seems a bit cleaner than tossing the macro onto everything, albeit slightly more obfuscated.
* Fix i386 atanhl (sNaN) (bug 20219).Joseph Myers2016-06-071-0/+2
| | | | | | | | | | | | | The i386 version of atanhl returns sNaN for sNaN input. This patch fixes it to add NaN arguments to themselves so it returns qNaN in this case. Tested for x86_64 and x86. [BZ #20219] * sysdeps/i386/fpu/e_atanhl.S (__ieee754_atanhl): Add NaN argument to itself. * math/libm-test.inc (atanh_test_data): Add sNaN tests.
* Fix i386 asinhl (sNaN) (bug 20218).Joseph Myers2016-06-071-0/+2
| | | | | | | | | | | | | | The i386 version of asinhl returns sNaN (without raising any exceptions) for sNaN input. This patch fixes it to add non-finite arguments to themselves, so that "invalid" is raised and qNaN returned. Tested for x86_64 and x86. [BZ #20218] * sysdeps/i386/fpu/s_asinhl.S (__asinhl): Add non-finite argument to itself. * math/libm-test.inc (asinh_test_data): Add sNaN tests.
* Fix dbl-64 asin (sNaN) (bug 20213).Joseph Myers2016-06-061-0/+2
| | | | | | | | | | | | | The dbl-64 version of asin returns sNaN for sNaN arguments. This patch fixes it to add NaN arguments to themselves so that qNaN is returned in this case. Tested for x86_64 and x86. [BZ #20213] * sysdeps/ieee754/dbl-64/e_asin.c (__ieee754_asin): Add NaN argument to itself. * math/libm-test.inc (asin_test_data): Add sNaN tests.
* Fix dbl-64 acos (sNaN) (bug 20212).Joseph Myers2016-06-061-0/+2
| | | | | | | | | | | | | The dbl-64 version of acos returns sNaN for sNaN arguments. This patch fixes it to add NaN arguments to themselves so that qNaN is returned in this case. Tested for x86_64 and x86. [BZ #20212] * sysdeps/ieee754/dbl-64/e_asin.c (__ieee754_acos): Add NaN argument to itself. * math/libm-test.inc (acos_test_data): Add sNaN tests.
* Fix x86/x86_64 nextafterl incrementing negative subnormals (bug 20205).Joseph Myers2016-06-031-0/+4
| | | | | | | | | | | | | | | | | The x86 / x86_64 implementation of nextafterl (also used for nexttowardl) produces incorrect results (NaNs) when negative subnormals, the low 32 bits of whose mantissa are zero, are incremented towards zero. This patch fixes this by disabling the logic to decrement the exponent in that case. Tested for x86_64 and x86. [BZ #20205] * sysdeps/i386/fpu/s_nextafterl.c (__nextafterl): Do not adjust exponent when incrementing negative subnormal with low mantissa word zero. * math/libm-test.inc (nextafter_test_data) [TEST_COND_intel96]: Add another test.
* Replace M_El with lit_e in libm-test.incPaul E. Murphy2016-05-271-3/+8
| | | | | This is useful in situations where the long double type is less precise than the type under test.
* Replace M_PI_4l with lit_pi_4_d in libm-test.incPaul E. Murphy2016-05-271-26/+28
| | | | | This is useful in situations where the long double type is less precise than the type under test.
* Replace M_PIl with lit_pi in libm-test.incPaul E. Murphy2016-05-271-38/+40
| | | | | This is useful in situations where the long double type is less precise than the type under test.
* Replace M_PI2l with lit_pi_2_d in libm-test.incPaul E. Murphy2016-05-274-157/+163
| | | | | | | This is useful in situations where the long double type is less precise than the type under test. This adds a new wrapper macro LITM(x) to each type to append the proper suffix onto macro constants found in math.h.
* Refactor M_ macros defined in libm-test.incPaul E. Murphy2016-05-271-33/+41
| | | | | | | | | These are local to the test suite. Rename them as a macro starting with lit_pi and a series of postfix operations to give us a constant starting with lit_pi. The lit prefix is intended to enable easy substitutions via gen-test-libm.pl if needed.
* Remove unused macros from libm-test.inc.Joseph Myers2016-05-271-25/+0
| | | | | | | | | | | | | | | | | | This patch removes various no-longer-used macros from libm-test.inc. NO_TEST_INLINE_FLOAT, NO_TEST_INLINE_DOUBLE and M_PI_6l would have been used before relevant tests were moved to auto-libm-test-in. TEST_COND_x86_64 and TEST_COND_x86 were for tests in auto-libm-test-in XFAILed for x86, and are no longer relevant now the bugs in question have been fixed and the XFAILing removed (if future x86-specific XFAILs become needed, they can always be added back). Tested for x86_64 and x86. * math/libm-test.inc (NO_TEST_INLINE_FLOAT): Remove macro. (NO_TEST_INLINE_DOUBLE): Likewise. (TEST_COND_x86_64): Likewise. (TEST_COND_x86): Likewise. (M_PI_6l): Likewise.
generated by cgit-pink 1.4.1 (git 2.36.1) at 2024-12-14 17:20:44 +0000