[PowerPC] Mark llvm.tan on vectors as expand

[ecnelises]

No description provided.

[llvmbot]

@llvm/pr-subscribers-backend-powerpc

Author: Qiu Chaofan (ecnelises)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/95507.diff

2 Files Affected:

• (modified) llvm/lib/Target/PowerPC/PPCISelLowering.cpp (+1)
• (modified) llvm/test/CodeGen/PowerPC/lower-intrinsics-nofast-mass.ll (+9)

diff --git a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
index 12e33ddb8eb53..e922b5d445e68 100644
--- a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -860,6 +860,7 @@ PPCTargetLowering::PPCTargetLowering(const PPCTargetMachine &TM,
       setOperationAction(ISD::FEXP2, VT, Expand);
       setOperationAction(ISD::FSIN, VT, Expand);
       setOperationAction(ISD::FCOS, VT, Expand);
+      setOperationAction(ISD::FTAN, VT, Expand);
       setOperationAction(ISD::FABS, VT, Expand);
       setOperationAction(ISD::FFLOOR, VT, Expand);
       setOperationAction(ISD::FCEIL,  VT, Expand);
diff --git a/llvm/test/CodeGen/PowerPC/lower-intrinsics-nofast-mass.ll b/llvm/test/CodeGen/PowerPC/lower-intrinsics-nofast-mass.ll
index 3d6e6da3b66d9..0c49d10e675a7 100644
--- a/llvm/test/CodeGen/PowerPC/lower-intrinsics-nofast-mass.ll
+++ b/llvm/test/CodeGen/PowerPC/lower-intrinsics-nofast-mass.ll
@@ -145,3 +145,12 @@ entry:
   %0 = tail call double @llvm.sin.f64(double %a)
   ret double %0
 }
+
+define <2 x double> @tan_v2f64_nofast(<2 x double> %a) {
+; CHECK-LABEL: tan_v2f64_nofast
+; CHECK-NOT: bl __xl_tan
+; CHECK: blr
+entry:
+  %0 = tail call <2 x double> @llvm.tan.v2f64(<2 x double> %a)
+  ret <2 x double> %0
+}

[farzonl]

Hi thanks for picking this up iI've been working on a general fix so I have a list of all the tests that need to be updated. For PowerPC it is:

• CodeGen/PowerPC/f128-arith.ll
• CodeGen/PowerPC/lower-intrinsics-afn-mass.ll
• CodeGen/PowerPC/lower-intrinsics-afn-mass_notail.ll
• CodeGen/PowerPC/lower-intrinsics-fast-mass.ll
• CodeGen/PowerPC/lower-intrinsics-fast-mass_notail.ll
• CodeGen/PowerPC/lower-intrinsics-nofast-mass.ll
• CodeGen/PowerPC/unal-altivec2.ll
• CodeGen/PowerPC/lower-intrinsics-fast-mass.ll
• CodeGen/PowerPC/lower-intrinsics-fast-mass_notail.ll
• CodeGen/PowerPC/lower-intrinsics-nofast-mass.ll
• CodeGen/PowerPC/unal-altivec2.ll

Also in llvm/include/llvm/Analysis/VecFuncs.def add

TLI_DEFINE_VECFUNC("llvm.tan.f64", "__tand2", FIXED(2), "_ZGV_LLVM_N2v") 
TLI_DEFINE_VECFUNC("llvm.tan.f32", "__tanf4", FIXED(4), "_ZGV_LLVM_N4v")

[chenzheng1030]

Do we need to handle STRICT_FTAN too?

[ecnelises]

It's already been handled by common logic, and there's already been test for constrained tan in vector-constrained-fp-intrinsics.ll

[ecnelises]

Hi @farzonl, do you mean this patch is not needed, because you will fix vector version of these math library nodes together?

[farzonl]

Hi @farzonl, do you mean this patch is not needed, because you will fix vector version of these math library nodes together?

My plan was to do an expand in TargetLoweringBase.cpp in this pr: #95518 That way I knock out more platforms at once.

I'm doing spot testing to make sure f16 and vector work. Also PowerPC needs llvm vectors to convert to IBM MASS Vector functions and my change doesn't do that either.