compiler: use the LLVM builtins everywhere

This gives some more optimization opportunities to LLVM, because it
understands these intrinsics. For example, it might convert
llvm.sqrt.f64 to llvm.sqrt.f32 if possible.
This commit is contained in:
Ayke van Laethem
2022-08-27 18:41:17 +02:00
committed by Ron Evans
parent ef912a132d
commit b8a6a1f62b
8 changed files with 387 additions and 248 deletions
+7 -5
View File
@@ -14,21 +14,23 @@ entry:
; Function Attrs: nounwind
define hidden double @main.mySqrt(double %x, i8* %context) unnamed_addr #1 {
entry:
%0 = call double @math.Sqrt(double %x, i8* undef) #2
%0 = call double @llvm.sqrt.f64(double %x)
ret double %0
}
declare double @math.Sqrt(double, i8*) #0
; Function Attrs: nofree nosync nounwind readnone speculatable willreturn
declare double @llvm.sqrt.f64(double) #2
; Function Attrs: nounwind
define hidden double @main.myTrunc(double %x, i8* %context) unnamed_addr #1 {
entry:
%0 = call double @math.Trunc(double %x, i8* undef) #2
%0 = call double @llvm.trunc.f64(double %x)
ret double %0
}
declare double @math.Trunc(double, i8*) #0
; Function Attrs: nofree nosync nounwind readnone speculatable willreturn
declare double @llvm.trunc.f64(double) #2
attributes #0 = { "target-features"="+armv7-m,+hwdiv,+soft-float,+strict-align,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
attributes #1 = { nounwind "target-features"="+armv7-m,+hwdiv,+soft-float,+strict-align,+thumb-mode,-aes,-bf16,-cdecp0,-cdecp1,-cdecp2,-cdecp3,-cdecp4,-cdecp5,-cdecp6,-cdecp7,-crc,-crypto,-d32,-dotprod,-dsp,-fp-armv8,-fp-armv8d16,-fp-armv8d16sp,-fp-armv8sp,-fp16,-fp16fml,-fp64,-fpregs,-fullfp16,-hwdiv-arm,-i8mm,-lob,-mve,-mve.fp,-neon,-pacbti,-ras,-sb,-sha2,-vfp2,-vfp2sp,-vfp3,-vfp3d16,-vfp3d16sp,-vfp3sp,-vfp4,-vfp4d16,-vfp4d16sp,-vfp4sp" }
attributes #2 = { nounwind }
attributes #2 = { nofree nosync nounwind readnone speculatable willreturn }