[PowerPC] Legalize saturating vector add/sub

[nemanjai]

Summary

These intrinsics and the corresponding ISD nodes were recently added. PPC has instructions that do this for vectors. Legalize them and add patterns to emit the satuarting instructions.

[lkail]

LGTM with one nit.

llvm/lib/Target/PowerPC/PPCInstrAltivec.td

871

Add // Saturates. before this line to separate following instructions from other classes of instructions?

[nemanjai]

llvm/lib/Target/PowerPC/PPCInstrAltivec.td

871

Sounds good.

[nemanjai]

Add a comment for the patterns.

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

PPCTargetLowering::PPCTargetLowering(const PPCTargetMachine &TM,
   if (Subtarget.hasVSX()) {
     setOperationAction(ISD::FMAXNUM_IEEE, MVT::f64, Legal);
     setOperationAction(ISD::FMAXNUM_IEEE, MVT::f32, Legal);
     setOperationAction(ISD::FMINNUM_IEEE, MVT::f64, Legal);
     setOperationAction(ISD::FMINNUM_IEEE, MVT::f32, Legal);
   }
 
   if (Subtarget.hasAltivec()) {
+    for (MVT VT : { MVT::v16i8, MVT::v8i16, MVT::v4i32 }) {
+      setOperationAction(ISD::SADDSAT, VT, Legal);
+      setOperationAction(ISD::SSUBSAT, VT, Legal);
+      setOperationAction(ISD::UADDSAT, VT, Legal);
+      setOperationAction(ISD::USUBSAT, VT, Legal);
+    }
     // First set operation action for all vector types to expand. Then we
     // will selectively turn on ones that can be effectively codegen'd.
     for (MVT VT : MVT::fixedlen_vector_valuetypes()) {
       // add/sub are legal for all supported vector VT's.
       setOperationAction(ISD::ADD, VT, Legal);
       setOperationAction(ISD::SUB, VT, Legal);
 
       // For v2i64, these are only valid with P8Vector. This is corrected after

llvm/lib/Target/PowerPC/PPCInstrAltivec.td

 
 // Rotates.
 def : Pat<(v16i8 (rotl v16i8:$vA, v16i8:$vB)),
           (v16i8 (VRLB v16i8:$vA, v16i8:$vB))>;
 def : Pat<(v8i16 (rotl v8i16:$vA, v8i16:$vB)),
           (v8i16 (VRLH v8i16:$vA, v8i16:$vB))>;
 def : Pat<(v4i32 (rotl v4i32:$vA, v4i32:$vB)),
           (v4i32 (VRLW v4i32:$vA, v4i32:$vB))>;
 
+// Saturating adds/subtracts.
+def : Pat<(v16i8 (saddsat v16i8:$vA, v16i8:$vB)), (v16i8 (VADDSBS $vA, $vB))>;
+def : Pat<(v16i8 (uaddsat v16i8:$vA, v16i8:$vB)), (v16i8 (VADDUBS $vA, $vB))>;
+def : Pat<(v8i16 (saddsat v8i16:$vA, v8i16:$vB)), (v8i16 (VADDSHS $vA, $vB))>;
+def : Pat<(v8i16 (uaddsat v8i16:$vA, v8i16:$vB)), (v8i16 (VADDUHS $vA, $vB))>;
+def : Pat<(v4i32 (saddsat v4i32:$vA, v4i32:$vB)), (v4i32 (VADDSWS $vA, $vB))>;
+def : Pat<(v4i32 (uaddsat v4i32:$vA, v4i32:$vB)), (v4i32 (VADDUWS $vA, $vB))>;
+def : Pat<(v16i8 (ssubsat v16i8:$vA, v16i8:$vB)), (v16i8 (VSUBSBS $vA, $vB))>;
+def : Pat<(v16i8 (usubsat v16i8:$vA, v16i8:$vB)), (v16i8 (VSUBUBS $vA, $vB))>;
+def : Pat<(v8i16 (ssubsat v8i16:$vA, v8i16:$vB)), (v8i16 (VSUBSHS $vA, $vB))>;
+def : Pat<(v8i16 (usubsat v8i16:$vA, v8i16:$vB)), (v8i16 (VSUBUHS $vA, $vB))>;
+def : Pat<(v4i32 (ssubsat v4i32:$vA, v4i32:$vB)), (v4i32 (VSUBSWS $vA, $vB))>;
+def : Pat<(v4i32 (usubsat v4i32:$vA, v4i32:$vB)), (v4i32 (VSUBUWS $vA, $vB))>;
+
 // Loads.
 def : Pat<(v4i32 (load xoaddr:$src)), (LVX xoaddr:$src)>;
 
 // Stores.
 def : Pat<(store v4i32:$rS, xoaddr:$dst),
           (STVX $rS, xoaddr:$dst)>;
 
 // Bit conversions.

llvm/test/CodeGen/PowerPC/saturating-intrinsics.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -O3 -mtriple=powerpc64le-unknown-unknown -ppc-asm-full-reg-names \
+; RUN:   -verify-machineinstrs -mcpu=pwr7 < %s | FileCheck %s
+define dso_local <16 x i8 > @vectorsaddb(<16 x i8 > %a, <16 x i8 > %b) {
+; CHECK-LABEL: vectorsaddb:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vaddsbs v2, v2, v3
+; CHECK-NEXT:    blr
+entry:
+  %call = call <16 x i8> @llvm.sadd.sat.v16i8(<16 x i8> %a, <16 x i8> %b)
+  ret <16 x i8> %call
+}
+
+define dso_local <16 x i8 > @vectorssubb(<16 x i8 > %a, <16 x i8 > %b) {
+; CHECK-LABEL: vectorssubb:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vsubsbs v2, v2, v3
+; CHECK-NEXT:    blr
+entry:
+  %call = call <16 x i8> @llvm.ssub.sat.v16i8(<16 x i8> %a, <16 x i8> %b)
+  ret <16 x i8> %call
+}
+
+define dso_local <16 x i8 > @vectoruaddb(<16 x i8 > %a, <16 x i8 > %b) {
+; CHECK-LABEL: vectoruaddb:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vaddubs v2, v2, v3
+; CHECK-NEXT:    blr
+entry:
+  %call = call <16 x i8> @llvm.uadd.sat.v16i8(<16 x i8> %a, <16 x i8> %b)
+  ret <16 x i8> %call
+}
+
+define dso_local <16 x i8 > @vectorusubb(<16 x i8 > %a, <16 x i8 > %b) {
+; CHECK-LABEL: vectorusubb:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vsububs v2, v2, v3
+; CHECK-NEXT:    blr
+entry:
+  %call = call <16 x i8> @llvm.usub.sat.v16i8(<16 x i8> %a, <16 x i8> %b)
+  ret <16 x i8> %call
+}
+
+define dso_local <8 x i16 > @vectorsaddh(<8 x i16 > %a, <8 x i16 > %b) {
+; CHECK-LABEL: vectorsaddh:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vaddshs v2, v2, v3
+; CHECK-NEXT:    blr
+entry:
+  %call = call <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16> %a, <8 x i16> %b)
+  ret <8 x i16> %call
+}
+
+define dso_local <8 x i16 > @vectorssubh(<8 x i16 > %a, <8 x i16 > %b) {
+; CHECK-LABEL: vectorssubh:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vsubshs v2, v2, v3
+; CHECK-NEXT:    blr
+entry:
+  %call = call <8 x i16> @llvm.ssub.sat.v8i16(<8 x i16> %a, <8 x i16> %b)
+  ret <8 x i16> %call
+}
+
+define dso_local <8 x i16 > @vectoruaddh(<8 x i16 > %a, <8 x i16 > %b) {
+; CHECK-LABEL: vectoruaddh:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vadduhs v2, v2, v3
+; CHECK-NEXT:    blr
+entry:
+  %call = call <8 x i16> @llvm.uadd.sat.v8i16(<8 x i16> %a, <8 x i16> %b)
+  ret <8 x i16> %call
+}
+
+define dso_local <8 x i16 > @vectorusubh(<8 x i16 > %a, <8 x i16 > %b) {
+; CHECK-LABEL: vectorusubh:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vsubuhs v2, v2, v3
+; CHECK-NEXT:    blr
+entry:
+  %call = call <8 x i16> @llvm.usub.sat.v8i16(<8 x i16> %a, <8 x i16> %b)
+  ret <8 x i16> %call
+}
+
+define dso_local <4 x i32 > @vectorsaddw(<4 x i32 > %a, <4 x i32 > %b) {
+; CHECK-LABEL: vectorsaddw:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vaddsws v2, v2, v3
+; CHECK-NEXT:    blr
+entry:
+  %call = call <4 x i32> @llvm.sadd.sat.v4i32(<4 x i32> %a, <4 x i32> %b)
+  ret <4 x i32> %call
+}
+
+define dso_local <4 x i32 > @vectorssubw(<4 x i32 > %a, <4 x i32 > %b) {
+; CHECK-LABEL: vectorssubw:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vsubsws v2, v2, v3
+; CHECK-NEXT:    blr
+entry:
+  %call = call <4 x i32> @llvm.ssub.sat.v4i32(<4 x i32> %a, <4 x i32> %b)
+  ret <4 x i32> %call
+}
+
+define dso_local <4 x i32 > @vectoruaddw(<4 x i32 > %a, <4 x i32 > %b) {
+; CHECK-LABEL: vectoruaddw:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vadduws v2, v2, v3
+; CHECK-NEXT:    blr
+entry:
+  %call = call <4 x i32> @llvm.uadd.sat.v4i32(<4 x i32> %a, <4 x i32> %b)
+  ret <4 x i32> %call
+}
+
+define dso_local <4 x i32 > @vectorusubw(<4 x i32 > %a, <4 x i32 > %b) {
+; CHECK-LABEL: vectorusubw:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    vsubuws v2, v2, v3
+; CHECK-NEXT:    blr
+entry:
+  %call = call <4 x i32> @llvm.usub.sat.v4i32(<4 x i32> %a, <4 x i32> %b)
+  ret <4 x i32> %call
+}
+
+declare <16 x i8> @llvm.sadd.sat.v16i8(<16 x i8>, <16 x i8>)
+declare <16 x i8> @llvm.ssub.sat.v16i8(<16 x i8>, <16 x i8>)
+declare <16 x i8> @llvm.uadd.sat.v16i8(<16 x i8>, <16 x i8>)
+declare <16 x i8> @llvm.usub.sat.v16i8(<16 x i8>, <16 x i8>)
+declare <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16>, <8 x i16>)
+declare <8 x i16> @llvm.ssub.sat.v8i16(<8 x i16>, <8 x i16>)
+declare <8 x i16> @llvm.uadd.sat.v8i16(<8 x i16>, <8 x i16>)
+declare <8 x i16> @llvm.usub.sat.v8i16(<8 x i16>, <8 x i16>)
+declare <4 x i32> @llvm.sadd.sat.v4i32(<4 x i32>, <4 x i32>)
+declare <4 x i32> @llvm.ssub.sat.v4i32(<4 x i32>, <4 x i32>)
+declare <4 x i32> @llvm.uadd.sat.v4i32(<4 x i32>, <4 x i32>)
+declare <4 x i32> @llvm.usub.sat.v4i32(<4 x i32>, <4 x i32>)