[GlobalISel][AArch64] Legalize G_ADD, G_SUB, G_AND, G_OR, and G_XOR f…

…or SVE (llvm#110561)

Credits: llvm#72976

LLVM ERROR: cannot select: %3:zpr(<vscale x 2 x s64>) = G_MUL %0:fpr,
%1:fpr (in function: xmulnxv2i64)

;; mul
define void @xmulnxv2i64(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b,
ptr %p) {
entry:
  %c = mul <vscale x 2 x i64> %a, %b
  store <vscale x 2 x i64> %c, ptr %p, align 16
   ret void
}

define void @mulnxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b,
ptr %p) {
entry:
  %c = mul <vscale x 4 x i32> %a, %b
  store <vscale x 4 x i32> %c, ptr %p, align 16
   ret void
}

define void @mulnxv8i16(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b,
ptr %p) {
entry:
  %c = mul <vscale x 8 x i16> %a, %b
  store <vscale x 8 x i16> %c, ptr %p, align 16
  ret void
}

define void @mulnxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b,
ptr %p) {
entry:
  %c = mul <vscale x 16 x i8> %a, %b
  store <vscale x 16 x i8> %c, ptr %p, align 16
  ret void
}

llvm/include/llvm/CodeGen/GlobalISel/LegalizerInfo.h

@@ -998,8 +998,7 @@ class LegalizeRuleSet {
         LegalizeAction::WidenScalar,
         [=](const LegalityQuery &Query) {
           const LLT VecTy = Query.Types[TypeIdx];
-          return VecTy.isVector() && !VecTy.isScalable() &&
-                 VecTy.getSizeInBits() < VectorSize;
+          return VecTy.isFixedVector() && VecTy.getSizeInBits() < VectorSize;
         },
         [=](const LegalityQuery &Query) {
           const LLT VecTy = Query.Types[TypeIdx];
@@ -1172,7 +1171,7 @@ class LegalizeRuleSet {
         LegalizeAction::MoreElements,
         [=](const LegalityQuery &Query) {
           LLT VecTy = Query.Types[TypeIdx];
-          return VecTy.isVector() && VecTy.getElementType() == EltTy &&
+          return VecTy.isFixedVector() && VecTy.getElementType() == EltTy &&
                  VecTy.getNumElements() < MinElements;
         },
         [=](const LegalityQuery &Query) {
@@ -1190,7 +1189,7 @@ class LegalizeRuleSet {
         LegalizeAction::MoreElements,
         [=](const LegalityQuery &Query) {
           LLT VecTy = Query.Types[TypeIdx];
-          return VecTy.isVector() && VecTy.getElementType() == EltTy &&
+          return VecTy.isFixedVector() && VecTy.getElementType() == EltTy &&
                  (VecTy.getNumElements() % NumElts != 0);
         },
         [=](const LegalityQuery &Query) {
@@ -1210,7 +1209,7 @@ class LegalizeRuleSet {
         LegalizeAction::FewerElements,
         [=](const LegalityQuery &Query) {
           LLT VecTy = Query.Types[TypeIdx];
-          return VecTy.isVector() && VecTy.getElementType() == EltTy &&
+          return VecTy.isFixedVector() && VecTy.getElementType() == EltTy &&
                  VecTy.getNumElements() > MaxElements;
         },
         [=](const LegalityQuery &Query) {
@@ -1231,6 +1230,9 @@ class LegalizeRuleSet {
     assert(MinTy.getElementType() == MaxTy.getElementType() &&
            "Expected element types to agree");
 
+    assert((!MinTy.isScalableVector() && !MaxTy.isScalableVector()) &&
+           "Unexpected scalable vectors");
+
     const LLT EltTy = MinTy.getElementType();
     return clampMinNumElements(TypeIdx, EltTy, MinTy.getNumElements())
         .clampMaxNumElements(TypeIdx, EltTy, MaxTy.getNumElements());

llvm/lib/Target/AArch64/AArch64GenRegisterBankInfo.def

@@ -183,6 +183,8 @@ unsigned AArch64GenRegisterBankInfo::getRegBankBaseIdxOffset(unsigned RBIdx,
     const unsigned MinSize = Size.getKnownMinValue();
     assert((!Size.isScalable() || MinSize >= 128) &&
            "Scalable vector types should have size of at least 128 bits");
+    if (Size.isScalable())
+      return 3;
     if (MinSize <= 16)
       return 0;
     if (MinSize <= 32)

llvm/lib/Target/AArch64/GISel/AArch64CallLowering.cpp

@@ -393,8 +393,8 @@ bool AArch64CallLowering::lowerReturn(MachineIRBuilder &MIRBuilder,
       // i1 is a special case because SDAG i1 true is naturally zero extended
       // when widened using ANYEXT. We need to do it explicitly here.
       auto &Flags = CurArgInfo.Flags[0];
-      if (MRI.getType(CurVReg).getSizeInBits() == 1 && !Flags.isSExt() &&
-          !Flags.isZExt()) {
+      if (MRI.getType(CurVReg).getSizeInBits() == TypeSize::getFixed(1) &&
+          !Flags.isSExt() && !Flags.isZExt()) {
         CurVReg = MIRBuilder.buildZExt(LLT::scalar(8), CurVReg).getReg(0);
       } else if (TLI.getNumRegistersForCallingConv(Ctx, CC, SplitEVTs[i]) ==
                  1) {

llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp

@@ -615,6 +615,7 @@ getMinClassForRegBank(const RegisterBank &RB, TypeSize SizeInBits,
   unsigned RegBankID = RB.getID();
 
   if (RegBankID == AArch64::GPRRegBankID) {
+    assert(!SizeInBits.isScalable() && "Unexpected scalable register size");
     if (SizeInBits <= 32)
       return GetAllRegSet ? &AArch64::GPR32allRegClass
                           : &AArch64::GPR32RegClass;
@@ -626,6 +627,12 @@ getMinClassForRegBank(const RegisterBank &RB, TypeSize SizeInBits,
   }
 
   if (RegBankID == AArch64::FPRRegBankID) {
+    if (SizeInBits.isScalable()) {
+      assert(SizeInBits == TypeSize::getScalable(128) &&
+             "Unexpected scalable register size");
+      return &AArch64::ZPRRegClass;
+    }
+
     switch (SizeInBits) {
     default:
       return nullptr;
@@ -964,7 +971,8 @@ getRegClassesForCopy(MachineInstr &I, const TargetInstrInfo &TII,
   // then we can pull it into the helpers that get the appropriate class for a
   // register bank. Or make a new helper that carries along some constraint
   // information.
-  if (SrcRegBank != DstRegBank && (DstSize == 1 && SrcSize == 1))
+  if (SrcRegBank != DstRegBank &&
+      (DstSize == TypeSize::getFixed(1) && SrcSize == TypeSize::getFixed(1)))
     SrcSize = DstSize = TypeSize::getFixed(32);
 
   return {getMinClassForRegBank(SrcRegBank, SrcSize, true),

llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp

@@ -91,6 +91,7 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
 
   const bool HasCSSC = ST.hasCSSC();
   const bool HasRCPC3 = ST.hasRCPC3();
+  const bool HasSVE = ST.hasSVE();
 
   getActionDefinitionsBuilder(
       {G_IMPLICIT_DEF, G_FREEZE, G_CONSTANT_FOLD_BARRIER})
@@ -127,7 +128,34 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
       .clampNumElements(0, v2s64, v2s64)
       .moreElementsToNextPow2(0);
 
-  getActionDefinitionsBuilder({G_ADD, G_SUB, G_MUL, G_AND, G_OR, G_XOR})
+  getActionDefinitionsBuilder({G_ADD, G_SUB, G_AND, G_OR, G_XOR})
+      .legalFor({s32, s64, v2s32, v2s64, v4s32, v4s16, v8s16, v16s8, v8s8})
+      .legalFor(HasSVE, {nxv16s8, nxv8s16, nxv4s32, nxv2s64})
+      .widenScalarToNextPow2(0)
+      .clampScalar(0, s32, s64)
+      .clampMaxNumElements(0, s8, 16)
+      .clampMaxNumElements(0, s16, 8)
+      .clampNumElements(0, v2s32, v4s32)
+      .clampNumElements(0, v2s64, v2s64)
+      .minScalarOrEltIf(
+          [=](const LegalityQuery &Query) {
+            return Query.Types[0].getNumElements() <= 2;
+          },
+          0, s32)
+      .minScalarOrEltIf(
+          [=](const LegalityQuery &Query) {
+            return Query.Types[0].getNumElements() <= 4;
+          },
+          0, s16)
+      .minScalarOrEltIf(
+          [=](const LegalityQuery &Query) {
+            return Query.Types[0].getNumElements() <= 16;
+          },
+          0, s8)
+      .scalarizeIf(scalarOrEltWiderThan(0, 64), 0)
+      .moreElementsToNextPow2(0);
+
+  getActionDefinitionsBuilder(G_MUL)
       .legalFor({s32, s64, v2s32, v2s64, v4s32, v4s16, v8s16, v16s8, v8s8})
       .widenScalarToNextPow2(0)
       .clampScalar(0, s32, s64)

llvm/test/CodeGen/AArch64/GlobalISel/legalizer-info-validation.mir

@@ -24,7 +24,6 @@
 # DEBUG-NEXT: .. the first uncovered imm index: 0, OK
 #
 # DEBUG-NEXT: G_MUL (opcode {{[0-9]+}}): 1 type index, 0 imm indices
-# DEBUG-NEXT: .. opcode {{[0-9]+}} is aliased to {{[0-9]+}}
 # DEBUG-NEXT: .. the first uncovered type index: 1, OK
 # DEBUG-NEXT: .. the first uncovered imm index: 0, OK
 #

llvm/test/CodeGen/AArch64/GlobalISel/sve-integer.ll

@@ -0,0 +1,208 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 3
+; RUN: llc < %s -mtriple aarch64 -mattr=+sve -aarch64-enable-gisel-sve=1  | FileCheck %s
+; RUN: llc < %s -mtriple aarch64 -mattr=+sve -global-isel -aarch64-enable-gisel-sve=1 | FileCheck %s
+
+;; add
+define <vscale x 2 x i64> @addnxv2i64(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: addnxv2i64:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    add z0.d, z0.d, z1.d
+; CHECK-NEXT:    ret
+entry:
+  %c = add <vscale x 2 x i64> %a, %b
+  ret <vscale x 2 x i64> %c
+}
+
+define <vscale x 4 x i32> @addnxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b) {
+; CHECK-LABEL: addnxv4i32:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    add z0.s, z0.s, z1.s
+; CHECK-NEXT:    ret
+entry:
+  %c = add <vscale x 4 x i32> %a, %b
+  ret <vscale x 4 x i32> %c
+}
+
+define <vscale x 8 x i16> @addnxv8i16(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b, ptr %p) {
+; CHECK-LABEL: addnxv8i16:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    add z0.h, z0.h, z1.h
+; CHECK-NEXT:    ret
+entry:
+  %c = add <vscale x 8 x i16> %a, %b
+  ret <vscale x 8 x i16> %c
+}
+
+define <vscale x 16 x i8> @addnxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b) {
+; CHECK-LABEL: addnxv16i8:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    add z0.b, z0.b, z1.b
+; CHECK-NEXT:    ret
+entry:
+  %c = add <vscale x 16 x i8> %a, %b
+  ret <vscale x 16 x i8> %c
+}
+
+;; sub
+define <vscale x 2 x i64> @subnxv2i64(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: subnxv2i64:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    sub z0.d, z0.d, z1.d
+; CHECK-NEXT:    ret
+entry:
+  %c = sub <vscale x 2 x i64> %a, %b
+  ret <vscale x 2 x i64> %c
+}
+
+define <vscale x 4 x i32> @subnxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b) {
+; CHECK-LABEL: subnxv4i32:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    sub z0.s, z0.s, z1.s
+; CHECK-NEXT:    ret
+entry:
+  %c = sub <vscale x 4 x i32> %a, %b
+  ret <vscale x 4 x i32> %c
+}
+
+define <vscale x 8 x i16> @subnxv8i16(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b, ptr %p) {
+; CHECK-LABEL: subnxv8i16:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    sub z0.h, z0.h, z1.h
+; CHECK-NEXT:    ret
+entry:
+  %c = sub <vscale x 8 x i16> %a, %b
+  ret <vscale x 8 x i16> %c
+}
+
+define <vscale x 16 x i8> @subnxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b) {
+; CHECK-LABEL: subnxv16i8:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    sub z0.b, z0.b, z1.b
+; CHECK-NEXT:    ret
+entry:
+  %c = sub <vscale x 16 x i8> %a, %b
+  ret <vscale x 16 x i8> %c
+}
+
+;; and
+define <vscale x 2 x i64> @andnxv2i64(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: andnxv2i64:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    and z0.d, z0.d, z1.d
+; CHECK-NEXT:    ret
+entry:
+  %c = and <vscale x 2 x i64> %a, %b
+  ret <vscale x 2 x i64> %c
+}
+
+define <vscale x 4 x i32> @andnxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b) {
+; CHECK-LABEL: andnxv4i32:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    and z0.d, z0.d, z1.d
+; CHECK-NEXT:    ret
+entry:
+  %c = and <vscale x 4 x i32> %a, %b
+  ret <vscale x 4 x i32> %c
+}
+
+define <vscale x 8 x i16> @andnxv8i16(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b, ptr %p) {
+; CHECK-LABEL: andnxv8i16:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    and z0.d, z0.d, z1.d
+; CHECK-NEXT:    ret
+entry:
+  %c = and <vscale x 8 x i16> %a, %b
+  ret <vscale x 8 x i16> %c
+}
+
+define <vscale x 16 x i8> @andnxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b) {
+; CHECK-LABEL: andnxv16i8:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    and z0.d, z0.d, z1.d
+; CHECK-NEXT:    ret
+entry:
+  %c = and <vscale x 16 x i8> %a, %b
+  ret <vscale x 16 x i8> %c
+}
+
+;; or
+define <vscale x 2 x i64> @ornxv2i64(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: ornxv2i64:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    orr z0.d, z0.d, z1.d
+; CHECK-NEXT:    ret
+entry:
+  %c = or <vscale x 2 x i64> %a, %b
+  ret <vscale x 2 x i64> %c
+}
+
+define <vscale x 4 x i32> @ornxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b) {
+; CHECK-LABEL: ornxv4i32:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    orr z0.d, z0.d, z1.d
+; CHECK-NEXT:    ret
+entry:
+  %c = or <vscale x 4 x i32> %a, %b
+  ret <vscale x 4 x i32> %c
+}
+
+define <vscale x 8 x i16> @ornxv8i16(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b, ptr %p) {
+; CHECK-LABEL: ornxv8i16:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    orr z0.d, z0.d, z1.d
+; CHECK-NEXT:    ret
+entry:
+  %c = or <vscale x 8 x i16> %a, %b
+  ret <vscale x 8 x i16> %c
+}
+
+define <vscale x 16 x i8> @ornxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b) {
+; CHECK-LABEL: ornxv16i8:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    orr z0.d, z0.d, z1.d
+; CHECK-NEXT:    ret
+entry:
+  %c = or <vscale x 16 x i8> %a, %b
+  ret <vscale x 16 x i8> %c
+}
+
+;; xor
+define <vscale x 2 x i64> @xornxv2i64(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b) {
+; CHECK-LABEL: xornxv2i64:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    eor z0.d, z0.d, z1.d
+; CHECK-NEXT:    ret
+entry:
+  %c = xor <vscale x 2 x i64> %a, %b
+  ret <vscale x 2 x i64> %c
+}
+
+define <vscale x 4 x i32> @xornxv4i32(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b) {
+; CHECK-LABEL: xornxv4i32:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    eor z0.d, z0.d, z1.d
+; CHECK-NEXT:    ret
+entry:
+  %c = xor <vscale x 4 x i32> %a, %b
+  ret <vscale x 4 x i32> %c
+}
+
+define <vscale x 8 x i16> @xornxv8i16(<vscale x 8 x i16> %a, <vscale x 8 x i16> %b, ptr %p) {
+; CHECK-LABEL: xornxv8i16:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    eor z0.d, z0.d, z1.d
+; CHECK-NEXT:    ret
+entry:
+  %c = xor <vscale x 8 x i16> %a, %b
+  ret <vscale x 8 x i16> %c
+}
+
+define <vscale x 16 x i8> @xornxv16i8(<vscale x 16 x i8> %a, <vscale x 16 x i8> %b) {
+; CHECK-LABEL: xornxv16i8:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    eor z0.d, z0.d, z1.d
+; CHECK-NEXT:    ret
+entry:
+  %c = xor <vscale x 16 x i8> %a, %b
+  ret <vscale x 16 x i8> %c
+}