[llvm] [AArch64][InstCombine] Canonicalize whilelo intrinsic (PR #151553)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-aarch64

Author: Matthew Devereau (MDevereau)

<details>
<summary>Changes</summary>

InstCombine llvm.aarch64.sve.whilelo to the generic LLVM llvm.get.active.lane.mask intrinsic

---
Full diff: https://github.com/llvm/llvm-project/pull/151553.diff


2 Files Affected:

- (modified) llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp (+11) 
- (added) llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-whilelo.ll (+66) 


``````````diff
diff --git a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
index 18ca22fc9f211..1220a0fc8ee82 100644
--- a/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
+++ b/llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp
@@ -2696,6 +2696,15 @@ static std::optional<Instruction *> instCombineDMB(InstCombiner &IC,
   return std::nullopt;
 }
 
+static std::optional<Instruction *> instCombineWhilelo(InstCombiner &IC,
+                                                       IntrinsicInst &II) {
+  return IC.replaceInstUsesWith(
+      II,
+      IC.Builder.CreateIntrinsic(Intrinsic::get_active_lane_mask,
+                                 {II.getType(), II.getOperand(0)->getType()},
+                                 {II.getOperand(0), II.getOperand(1)}));
+}
+
 static std::optional<Instruction *> instCombinePTrue(InstCombiner &IC,
                                                      IntrinsicInst &II) {
   if (match(II.getOperand(0), m_ConstantInt<AArch64SVEPredPattern::all>()))
@@ -2830,6 +2839,8 @@ AArch64TTIImpl::instCombineIntrinsic(InstCombiner &IC,
     return instCombineSVEDupqLane(IC, II);
   case Intrinsic::aarch64_sve_insr:
     return instCombineSVEInsr(IC, II);
+  case Intrinsic::aarch64_sve_whilelo:
+    return instCombineWhilelo(IC, II);
   case Intrinsic::aarch64_sve_ptrue:
     return instCombinePTrue(IC, II);
   case Intrinsic::aarch64_sve_uxtb:
diff --git a/llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-whilelo.ll b/llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-whilelo.ll
new file mode 100644
index 0000000000000..9dde171217432
--- /dev/null
+++ b/llvm/test/Transforms/InstCombine/AArch64/sve-intrinsic-whilelo.ll
@@ -0,0 +1,66 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S -passes=instcombine < %s | FileCheck %s
+
+target triple = "aarch64-unknown-linux-gnu"
+
+
+define <vscale x 4 x float> @const_whilelo_nxv4i32(ptr %0) #0 {
+; CHECK-LABEL: define <vscale x 4 x float> @const_whilelo_nxv4i32(
+; CHECK-SAME: ptr [[TMP0:%.*]]) {
+; CHECK-NEXT:    [[MASK:%.*]] = call <vscale x 4 x i1> @llvm.get.active.lane.mask.nxv4i1.i32(i32 0, i32 4)
+; CHECK-NEXT:    [[LOAD:%.*]] = tail call <vscale x 4 x float> @llvm.masked.load.nxv4f32.p0(ptr nonnull [[TMP0]], i32 1, <vscale x 4 x i1> [[MASK]], <vscale x 4 x float> zeroinitializer)
+; CHECK-NEXT:    ret <vscale x 4 x float> [[LOAD]]
+;
+  %mask = tail call <vscale x 4 x i1> @llvm.aarch64.sve.whilelo.nxv4i1.i32(i32 0, i32 4)
+  %load = tail call <vscale x 4 x float> @llvm.masked.load.nxv4f32.p0(ptr nonnull %0, i32 1, <vscale x 4 x i1> %mask, <vscale x 4 x float> zeroinitializer)
+  ret <vscale x 4 x float> %load
+}
+
+define <vscale x 8 x float> @const_whilelo_nxv8f32(ptr %0) #0 {
+; CHECK-LABEL: define <vscale x 8 x float> @const_whilelo_nxv8f32(
+; CHECK-SAME: ptr [[TMP0:%.*]]) {
+; CHECK-NEXT:    [[MASK:%.*]] = call <vscale x 8 x i1> @llvm.get.active.lane.mask.nxv8i1.i32(i32 0, i32 8)
+; CHECK-NEXT:    [[LOAD:%.*]] = tail call <vscale x 8 x float> @llvm.masked.load.nxv8f32.p0(ptr nonnull [[TMP0]], i32 1, <vscale x 8 x i1> [[MASK]], <vscale x 8 x float> zeroinitializer)
+; CHECK-NEXT:    ret <vscale x 8 x float> [[LOAD]]
+;
+  %mask = tail call <vscale x 8 x i1> @llvm.aarch64.sve.whilelo.nxv8i1.i32(i32 0, i32 8)
+  %load = tail call <vscale x 8 x float> @llvm.masked.load.nxv8f32.p0(ptr nonnull %0, i32 1, <vscale x 8 x i1> %mask, <vscale x 8 x float> zeroinitializer)
+  ret <vscale x 8 x float> %load
+}
+
+define <vscale x 8 x i16> @const_whilelo_nxv8i16(ptr %0) #0 {
+; CHECK-LABEL: define <vscale x 8 x i16> @const_whilelo_nxv8i16(
+; CHECK-SAME: ptr [[TMP0:%.*]]) {
+; CHECK-NEXT:    [[MASK:%.*]] = call <vscale x 8 x i1> @llvm.get.active.lane.mask.nxv8i1.i32(i32 0, i32 8)
+; CHECK-NEXT:    [[LOAD:%.*]] = tail call <vscale x 8 x i16> @llvm.masked.load.nxv8i16.p0(ptr nonnull [[TMP0]], i32 1, <vscale x 8 x i1> [[MASK]], <vscale x 8 x i16> zeroinitializer)
+; CHECK-NEXT:    ret <vscale x 8 x i16> [[LOAD]]
+;
+  %mask = tail call <vscale x 8 x i1> @llvm.aarch64.sve.whilelo.nxv8i1.i16(i32 0, i32 8)
+  %load = tail call <vscale x 8 x i16> @llvm.masked.load.nxv8i16.p0(ptr nonnull %0, i32 1, <vscale x 8 x i1> %mask, <vscale x 8 x i16> zeroinitializer)
+  ret <vscale x 8 x i16> %load
+}
+
+define <vscale x 16 x i8> @const_whilelo_nxv16i8(ptr %0) #0 {
+; CHECK-LABEL: define <vscale x 16 x i8> @const_whilelo_nxv16i8(
+; CHECK-SAME: ptr [[TMP0:%.*]]) {
+; CHECK-NEXT:    [[MASK:%.*]] = call <vscale x 16 x i1> @llvm.get.active.lane.mask.nxv16i1.i32(i32 0, i32 16)
+; CHECK-NEXT:    [[LOAD:%.*]] = tail call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0(ptr nonnull [[TMP0]], i32 1, <vscale x 16 x i1> [[MASK]], <vscale x 16 x i8> zeroinitializer)
+; CHECK-NEXT:    ret <vscale x 16 x i8> [[LOAD]]
+;
+  %mask = tail call <vscale x 16 x i1> @llvm.aarch64.sve.whilelo.nxv16i1.i8(i32 0, i32 16)
+  %load = tail call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0(ptr nonnull %0, i32 1, <vscale x 16 x i1> %mask, <vscale x 16 x i8> zeroinitializer)
+  ret <vscale x 16 x i8> %load
+}
+
+
+define <vscale x 16 x i8> @whilelo_nxv16i8(ptr %0, i32 %a, i32 %b) #0 {
+; CHECK-LABEL: define <vscale x 16 x i8> @whilelo_nxv16i8(
+; CHECK-SAME: ptr [[TMP0:%.*]], i32 [[A:%.*]], i32 [[B:%.*]]) {
+; CHECK-NEXT:    [[MASK:%.*]] = call <vscale x 16 x i1> @llvm.get.active.lane.mask.nxv16i1.i32(i32 [[A]], i32 [[B]])
+; CHECK-NEXT:    [[LOAD:%.*]] = tail call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0(ptr nonnull [[TMP0]], i32 1, <vscale x 16 x i1> [[MASK]], <vscale x 16 x i8> zeroinitializer)
+; CHECK-NEXT:    ret <vscale x 16 x i8> [[LOAD]]
+;
+  %mask = tail call <vscale x 16 x i1> @llvm.aarch64.sve.whilelo.nxv16i1.i8(i32 %a, i32 %b)
+  %load = tail call <vscale x 16 x i8> @llvm.masked.load.nxv16i8.p0(ptr nonnull %0, i32 1, <vscale x 16 x i1> %mask, <vscale x 16 x i8> zeroinitializer)
+  ret <vscale x 16 x i8> %load
+}

``````````

</details>


https://github.com/llvm/llvm-project/pull/151553