Hello.
This issue is related to
https://go-review.googlesource.com/c/gollvm/+/274574
.
I think I have some misunderstanding on how you used to deal with CPU
models, for LLVM.
First things first - I had success with using
#include "llvm/ADT/StringRef.h"
#include "llvm/ADT/StringMap.h"
#include "llvm/MC/SubtargetFeature.h"
#include "llvm/Support/Host.h"
using namespace llvm;
SubtargetFeatures Features1;
int main (int argc, char **argv)
{
sys::getHostCPUName();
StringMap HostFeatures;
if (sys::getHostCPUFeatures(HostFeatures))
for (auto &F : HostFeatures)
Features1.AddFeature(F.first(), F.second);
printf("test %s", Features1.getString().c_str());
printf("\nsomething else\n");
return 0;
}
. It gives me such a set of CPU features:
+sse2,-tsxldtrk,-cx16,-sahf,-tbm,-avx512ifma,-sha,-gfni,-fma4,-vpclmulqdq,-prfchw,-bmi2,-cldemote,-fsgsbase,-ptwrite,-amx-tile,-avx512bf16,-popcnt,-aes,-avx512bitalg,-movdiri,-xsaves,-avx512er,-xsavec,-avx512vnni,-amx-bf16,-avx512vpopcntdq,-pconfig,-clwb,-avx512f,-clzero,-pku,+mmx,-lwp,-rdpid,-xop,-rdseed,-waitpkg,-movdir64b,-sse4a,-avx512bw,-clflushopt,-xsave,-avx512vbmi2,-64bit,-avx512vl,-serialize,-invpcid,-avx512cd,-avx,-vaes,+cx8,-fma,-rtm,-bmi,-enqcmd,-rdrnd,-mwaitx,-sse4.1,-sse4.2,-avx2,+fxsr,-wbnoinvd,+sse,-lzcnt,-pclmul,-prefetchwt1,-f16c,-ssse3,-sgx,-shstk,+cmov,-avx512vbmi,-amx-int8,-movbe,-avx512vp2intersect,-xsaveopt,-avx512dq,-adx,-avx512pf,+sse3
$ llc --version
provides
Default target: i686-pc-linux-gnu
Host CPU: yonah
.
I tried to update the capture-fcn-attributes.go file, like this:
var supportedTriples []string = []string{
"x86_64-unknown-linux-gnu",
"i686-pc-linux-gnu",
"aarch64-unknown-linux-gnu",
}
.
When I tried the generator
capture-fcn-attributes -o /tmp/cpu_feature_list
it generated me a broad list.
The header contained
Ubuntu clang version 11.0.0-++20200721055954+cebd637c886-1exp1
20200721161335.13
.
I found
// triple: i686-pc-linux-gnu
static const CpuAttrs attrs1[] = {
// first entry is default cpu
{ "i686", "+cx8,+x87"},
and (inside the hashmap)
{ "yonah", "+cx8,+fxsr,+mmx,+sse,+sse2,+sse3,+x87"},
, which is not what I have supported (for Intel Celeron M440).
Clang reports "unsupported CPU features" on any non-provided one.
So that is one big problem.
Next problem is that
const TripleCpus triples[] = {
{ "x86_64-unknown-linux-gnu", &attrs0[0] },
{ "i686-pc-linux-gnu", &attrs1[0] },
{ "aarch64-unknown-linux-gnu", &attrs2[0] },
{ "", nullptr } // sentinel
};
is not targeting to yonah, while llc is targeting it.
It is always some "default" CPU model and, in fact, your code never
provided extraction of the CPU model (from llc).
To make my observation complete - I am providing what is generated via
capture-fcn-attributes -cpu yonah
:
// triple: x86_64-unknown-linux-gnu
static const CpuAttrs attrs0[] = {
// first entry is default cpu
{ "x86-64", "+cx8,+fxsr,+mmx,+sse,+sse2,+x87"},
{ "", "" } // sentinel
};
// triple: i686-pc-linux-gnu
static const CpuAttrs attrs1[] = {
// first entry is default cpu
{ "i686", "+cx8,+x87"},
{ "yonah", "+cx8,+fxsr,+mmx,+sse,+sse2,+sse3,+x87"},
{ "", "" } // sentinel
};
// triple: aarch64-unknown-linux-gnu
static const CpuAttrs attrs2[] = {
// first entry is default cpu
{ "generic", "+neon"},
{ "", "" } // sentinel
};
const TripleCpus triples[] = {
{ "x86_64-unknown-linux-gnu", &attrs0[0] },
{ "i686-pc-linux-gnu", &attrs1[0] },
{ "aarch64-unknown-linux-gnu", &attrs2[0] },
{ "", nullptr } // sentinel
};
.
I tried
capture-fcn-attributes -cpu yonah -triples i686-pc-linux-gnu
and got
// triple: i686-pc-linux-gnu
static const CpuAttrs attrs0[] = {
// first entry is default cpu
{ "i686", "+cx8,+x87"},
{ "yonah", "+cx8,+fxsr,+mmx,+sse,+sse2,+sse3,+x87"},
{ "", "" } // sentinel
};
const TripleCpus triples[] = {
{ "i686-pc-linux-gnu", &attrs0[0] },
{ "", nullptr } // sentinel
};
.
I understand that your strategy worked find on Intel based system-on-board
machines - but didn't try something for AMD (yet).
Nevertheless I have these issues on i686 - so I am proposing to perform a
review.
Ivan
--
You received this message because you are subscribed to the Google Groups
"golang-nuts" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to golang-nuts+unsubscr...@googlegroups.com.
To view this discussion on the web visit
https://groups.google.com/d/msgid/golang-nuts/aac8c576-9763-4bba-96a1-51f545084630n%40googlegroups.com.
