[Bug jit/107877] New: segfault in libgccjit when using asse
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107877 Bug ID: 107877 Summary: segfault in libgccjit when using asse Product: gcc Version: 11.3.0 Status: UNCONFIRMED Severity: normal Priority: P3 Component: jit Assignee: dmalcolm at gcc dot gnu.org Reporter: andreas_roever at web dot de Target Milestone: --- Created attachment 53969 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=53969=edit example of bug due to bug 107849 I tried to create functions that emulate the bultin AVX intrinsics. I seems like there is a bug in libjit though. In the attached sample there is a simpele function test at the top that should provides functionality identical to the _mm256_broadcast_ss intrinsic. When I try to re-create this function inside the jit compiler. It simply crashes with a segfault inside the compile function. Am I doing something wrong that causes the segfault? Also I guess an error message would be more helpful than a segfault :)
[Bug jit/107849] New: All SIMD instrinsics are missing
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107849 Bug ID: 107849 Summary: All SIMD instrinsics are missing Product: gcc Version: 11.3.0 Status: UNCONFIRMED Severity: normal Priority: P3 Component: jit Assignee: dmalcolm at gcc dot gnu.org Reporter: andreas_roever at web dot de Target Milestone: --- Created attachment 53956 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=53956=edit A simple example showing the problem It seems like the jit library only provies builtin functions for the "standard" builtin functions. All the i386 SSE, AVX, ... stuff is missing. I attache a silly little program to show what I mean. The program defines a jit function that takes an argument, calculates the sqrt using the builtin function and returns that value. It als calls the builtin directly and then prints out the results When compiled without defining VECTOR it uses a single float and everything works as expected. When VECTOR is defines everything gets redefined to use an 256bit vector with 8 floats and also uses the appropriate builtin function. In that case the jit function doesn't find the builtin. I guess that that is the case because these builtins are not defined in the same .def file as the functions that are known by libgccjit. The library only uses the builtins defined in the gcc subdirectoy, but the SIMD builtins are defined in the folder gcc/config/i386
[Bug libstdc++/103005] experimental simd sin and cos with big arguments returns values bigger than 1
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103005 --- Comment #3 from Andreas Röver --- I think the source of the problem is function __fold_input(const simd& __x) in simd_math.h this function should return values between -pi/4 and pi/4 but doesn't for the big argument. E.g for 2985064393126969344 it returns 27.678747 according to wolfram alpha it should return 0.189812 This "big" value throws off the calculation of the power series that is performed after this the 27.678646 is the result of a rounding error 2985064393126969344 / (pi/2) is (according to wolfram alpha again) 1900351014455063569.620838 but the CPU gets 1900351014455063552 This seems to be the closest value that the CPU can get to the real value. the difference between those 2 times pi/2 is the above mentioned 27.67... ... now my numerics skills are not good enough to find a cure for that problem.
[Bug libstdc++/103005] experimental simd sin and cos with big arguments returns values bigger than 1
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103005 --- Comment #2 from Andreas Röver --- I think the source of the problem is function __fold_input(const simd& __x) in simd_math.h this function should return values between -pi/4 and pi/4 but doesn't for the big argument. E.g for 2985064393126969344 it returns 27.678747 according to wolfram alpha it should return 0.189812 This "big" value throws off the calculation of the power series that is performed after this the 27.678646 is the result of a rounding error 2985064393126969344 / (pi/2) is (according to wolfram alpha again) 1900351014455063569.620838 but the CPU gets 1900351014455063552 This seems to be the closest value that the CPU can get to the real value. the difference between those 2 times pi/2 is the above mentioned 27.67... ... now my numerics skills are not good enough to find a cure for that problem.
[Bug libstdc++/103005] New: experimental simd sin and cos with big arguments returns values bigger than 1
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103005 Bug ID: 103005 Summary: experimental simd sin and cos with big arguments returns values bigger than 1 Product: gcc Version: 11.2.0 Status: UNCONFIRMED Severity: normal Priority: P3 Component: libstdc++ Assignee: unassigned at gcc dot gnu.org Reporter: andreas_roever at web dot de Target Milestone: --- Created attachment 51704 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=51704=edit example program displaying the problem there seems to be a bug in the sin and cos functions for the experimental/simd header. When giving a very big argument to those functions they return values outside the range -1 .. 1. I understand that the result will be quite meaningless for such big values. But I'd argue that it should always be in the range -1..1. so calculating sin(-2985064393126969344) gives 2527133379.389218 This is on Linux compiled with g++-11.2.0 -g -march=native --std=c++20 test1.cpp here the CPU info (for one core) processor : 0 vendor_id : GenuineIntel cpu family : 6 model : 94 model name : Intel(R) Core(TM) i5-6500 CPU @ 3.20GHz stepping: 3 microcode : 0xc2 cpu MHz : 3301.335 cache size : 6144 KB physical id : 0 siblings: 4 core id : 0 cpu cores : 4 apicid : 0 initial apicid : 0 fpu : yes fpu_exception : yes cpuid level : 22 wp : yes flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault invpcid_single pti ibrs ibpb stibp tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 hle avx2 smep bmi2 erms invpcid rtm mpx rdseed adx smap clflushopt intel_pt xsaveopt xsavec xgetbv1 xsaves dtherm ida arat pln pts hwp hwp_notify hwp_act_window hwp_epp vmx flags : vnmi preemption_timer invvpid ept_x_only ept_ad ept_1gb flexpriority tsc_offset vtpr mtf vapic ept vpid unrestricted_guest ple shadow_vmcs pml bugs: cpu_meltdown spectre_v1 spectre_v2 spec_store_bypass l1tf mds swapgs taa itlb_multihit srbds bogomips: 6399.96 clflush size: 64 cache_alignment : 64 address sizes : 39 bits physical, 48 bits virtual