Skip to content

[GlobalISel][AArch64] Legalize G_INSERT_VECTOR_ELT for SVE #114470

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 3 commits into from
Nov 1, 2024
Merged

[GlobalISel][AArch64] Legalize G_INSERT_VECTOR_ELT for SVE #114470

merged 3 commits into from
Nov 1, 2024

Conversation

tschuett
Copy link

@tschuett tschuett commented Oct 31, 2024 •
edited
Loading

There are patterns for:

  • {nxv2s32, s32, s64},
  • {nxv4s16, s16, s64},
  • {nxv2s16, s16, s64}

@tschuett
[GlobalISel][AArch64] Legalize G_INSERT_VECTOR_ELT for SVE
f779656 
There are patterns for:
* {nxv2s32, s32, s64},
* {nxv4s16, s16, s64},
* {nxv2s16, s16, s64}

For {nxv16s8, s32, s64}, {nxv8s16, s32, s64} element type and vector element type differ,
which makes LLVM-IR unhappy.
@tschuett tschuett requested review from arsenm and davemgreen October 31, 2024 21:47
@llvmbot
Copy link
Member

llvmbot commented Oct 31, 2024

@llvm/pr-subscribers-backend-aarch64

Author: Thorsten Schütt (tschuett)

Changes

There are patterns for:

  • {nxv2s32, s32, s64},
  • {nxv4s16, s16, s64},
  • {nxv2s16, s16, s64}

For {nxv16s8, s32, s64}, {nxv8s16, s32, s64} element type and vector element type differ, which makes LLVM-IR unhappy.

Full diff: https://github.com/llvm/llvm-project/pull/114470.diff

5 Files Affected:

  • (modified) llvm/include/llvm/CodeGen/GlobalISel/LegalizerInfo.h (+20)
  • (modified) llvm/lib/CodeGen/GlobalISel/LegalityPredicates.cpp (+11)
  • (modified) llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp (+1)
  • (modified) llvm/lib/Target/AArch64/GISel/AArch64PostLegalizerLowering.cpp (+3)
  • (modified) llvm/test/CodeGen/AArch64/extract-vector-elt.ll (+76-8)
diff --git a/llvm/include/llvm/CodeGen/GlobalISel/LegalizerInfo.h b/llvm/include/llvm/CodeGen/GlobalISel/LegalizerInfo.h
index 6d71c150c8da6b..343a0172ff39ed 100644
--- a/llvm/include/llvm/CodeGen/GlobalISel/LegalizerInfo.h
+++ b/llvm/include/llvm/CodeGen/GlobalISel/LegalizerInfo.h
@@ -273,6 +273,11 @@ inline LegalityPredicate typeIsNot(unsigned TypeIdx, LLT Type) {
 LegalityPredicate
 typePairInSet(unsigned TypeIdx0, unsigned TypeIdx1,
               std::initializer_list<std::pair<LLT, LLT>> TypesInit);
+/// True iff the given types for the given tuple of type indexes is one of the
+/// specified type tuple.
+LegalityPredicate
+typeTupleInSet(unsigned TypeIdx0, unsigned TypeIdx1, unsigned Type2,
+               std::initializer_list<std::tuple<LLT, LLT, LLT>> TypesInit);
 /// True iff the given types for the given pair of type indexes is one of the
 /// specified type pairs.
 LegalityPredicate typePairAndMemDescInSet(
@@ -504,6 +509,15 @@ class LegalizeRuleSet {
     using namespace LegalityPredicates;
     return actionIf(Action, typePairInSet(typeIdx(0), typeIdx(1), Types));
   }
+
+  LegalizeRuleSet &
+  actionFor(LegalizeAction Action,
+            std::initializer_list<std::tuple<LLT, LLT, LLT>> Types) {
+    using namespace LegalityPredicates;
+    return actionIf(Action,
+                    typeTupleInSet(typeIdx(0), typeIdx(1), typeIdx(2), Types));
+  }
+
   /// Use the given action when type indexes 0 and 1 is any type pair in the
   /// given list.
   /// Action should be an action that requires mutation.
@@ -615,6 +629,12 @@ class LegalizeRuleSet {
       return *this;
     return actionFor(LegalizeAction::Legal, Types);
   }
+  LegalizeRuleSet &
+  legalFor(bool Pred, std::initializer_list<std::tuple<LLT, LLT, LLT>> Types) {
+    if (!Pred)
+      return *this;
+    return actionFor(LegalizeAction::Legal, Types);
+  }
   /// The instruction is legal when type index 0 is any type in the given list
   /// and imm index 0 is anything.
   LegalizeRuleSet &legalForTypeWithAnyImm(std::initializer_list<LLT> Types) {
diff --git a/llvm/lib/CodeGen/GlobalISel/LegalityPredicates.cpp b/llvm/lib/CodeGen/GlobalISel/LegalityPredicates.cpp
index 8fe48195c610be..dc7ed6cbe8b7da 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegalityPredicates.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegalityPredicates.cpp
@@ -49,6 +49,17 @@ LegalityPredicate LegalityPredicates::typePairInSet(
   };
 }
 
+LegalityPredicate LegalityPredicates::typeTupleInSet(
+    unsigned TypeIdx0, unsigned TypeIdx1, unsigned TypeIdx2,
+    std::initializer_list<std::tuple<LLT, LLT, LLT>> TypesInit) {
+  SmallVector<std::tuple<LLT, LLT, LLT>, 4> Types = TypesInit;
+  return [=](const LegalityQuery &Query) {
+    std::tuple<LLT, LLT, LLT> Match = {
+        Query.Types[TypeIdx0], Query.Types[TypeIdx1], Query.Types[TypeIdx2]};
+    return llvm::is_contained(Types, Match);
+  };
+}
+
 LegalityPredicate LegalityPredicates::typePairAndMemDescInSet(
     unsigned TypeIdx0, unsigned TypeIdx1, unsigned MMOIdx,
     std::initializer_list<TypePairAndMemDesc> TypesAndMemDescInit) {
diff --git a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
index 400024922124cd..31627060c9a6bc 100644
--- a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
+++ b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
@@ -978,6 +978,7 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
   getActionDefinitionsBuilder(G_INSERT_VECTOR_ELT)
       .legalIf(
           typeInSet(0, {v16s8, v8s8, v8s16, v4s16, v4s32, v2s32, v2s64, v2p0}))
+      .legalFor(HasSVE, {{nxv4s32, s32, s64}, {nxv2s64, s64, s64}})
       .moreElementsToNextPow2(0)
       .widenVectorEltsToVectorMinSize(0, 64)
       .clampNumElements(0, v8s8, v16s8)
diff --git a/llvm/lib/Target/AArch64/GISel/AArch64PostLegalizerLowering.cpp b/llvm/lib/Target/AArch64/GISel/AArch64PostLegalizerLowering.cpp
index b40fe55fdfaf67..2bcfdc1b46873b 100644
--- a/llvm/lib/Target/AArch64/GISel/AArch64PostLegalizerLowering.cpp
+++ b/llvm/lib/Target/AArch64/GISel/AArch64PostLegalizerLowering.cpp
@@ -426,6 +426,9 @@ void applyNonConstInsert(MachineInstr &MI, MachineRegisterInfo &MRI,
   LLT EltTy = MRI.getType(Insert.getElementReg());
   LLT IdxTy = MRI.getType(Insert.getIndexReg());
 
+  if (VecTy.isScalableVector())
+    return;
+
   // Create a stack slot and store the vector into it
   MachineFunction &MF = Builder.getMF();
   Align Alignment(
diff --git a/llvm/test/CodeGen/AArch64/extract-vector-elt.ll b/llvm/test/CodeGen/AArch64/extract-vector-elt.ll
index 0481d997d24faf..b7cc04fa6edc58 100644
--- a/llvm/test/CodeGen/AArch64/extract-vector-elt.ll
+++ b/llvm/test/CodeGen/AArch64/extract-vector-elt.ll
@@ -1,6 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
-; RUN: llc -mtriple=aarch64 -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-SD
-; RUN: llc -mtriple=aarch64 -global-isel -global-isel-abort=2 -verify-machineinstrs %s -o - 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-GI
+; RUN: llc -mtriple=aarch64 -mattr=+sve -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-SD
+; RUN: llc -mtriple=aarch64 -mattr=+sve -aarch64-enable-gisel-sve=1 -global-isel -global-isel-abort=2 -verify-machineinstrs %s -o - 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-GI
 
 ; CHECK-GI:       warning: Instruction selection used fallback path for extract_v4i32_vector_insert
 ; CHECK-GI-NEXT:  warning: Instruction selection used fallback path for extract_v4i32_vector_insert_const
@@ -143,11 +143,12 @@ define i64 @extract_v2i64_extract_build_vector_opaque(<2 x i64> %a, i32 %c) {
 ; CHECK-SD:       // %bb.0: // %entry
 ; CHECK-SD-NEXT:    sub sp, sp, #16
 ; CHECK-SD-NEXT:    .cfi_def_cfa_offset 16
-; CHECK-SD-NEXT:    adrp x8, .LCPI8_0
+; CHECK-SD-NEXT:    mov x8, #-31 // =0xffffffffffffffe1
 ; CHECK-SD-NEXT:    // kill: def $w0 killed $w0 def $x0
-; CHECK-SD-NEXT:    ldr q0, [x8, :lo12:.LCPI8_0]
+; CHECK-SD-NEXT:    index z0.d, #0, x8
 ; CHECK-SD-NEXT:    mov x8, sp
 ; CHECK-SD-NEXT:    bfi x8, x0, #3, #1
+; CHECK-SD-NEXT:    add z0.d, z0.d, #42 // =0x2a
 ; CHECK-SD-NEXT:    str q0, [sp]
 ; CHECK-SD-NEXT:    ldr x0, [x8]
 ; CHECK-SD-NEXT:    add sp, sp, #16
@@ -1043,11 +1044,10 @@ entry:
 define i32 @extract_v4i32_phi(i64 %val, i32  %limit, ptr %ptr) {
 ; CHECK-SD-LABEL: extract_v4i32_phi:
 ; CHECK-SD:       // %bb.0: // %entry
-; CHECK-SD-NEXT:    dup v1.2s, w0
-; CHECK-SD-NEXT:    adrp x8, .LCPI41_0
+; CHECK-SD-NEXT:    index z1.s, #1, #1
+; CHECK-SD-NEXT:    dup v2.2s, w0
 ; CHECK-SD-NEXT:    movi v0.2s, #16
-; CHECK-SD-NEXT:    ldr d2, [x8, :lo12:.LCPI41_0]
-; CHECK-SD-NEXT:    add v1.2s, v1.2s, v2.2s
+; CHECK-SD-NEXT:    add v1.2s, v2.2s, v1.2s
 ; CHECK-SD-NEXT:  .LBB41_1: // %loop
 ; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
 ; CHECK-SD-NEXT:    fmov w8, s1
@@ -1100,4 +1100,72 @@ ret:
   ret i32 %3
 }
 
+define <vscale x 2 x i64> @insert_vscale_2_i64_zero(<vscale x 2 x i64> %vec, i64 %elt) {
+; CHECK-SD-LABEL: insert_vscale_2_i64_zero:
+; CHECK-SD:       // %bb.0: // %entry
+; CHECK-SD-NEXT:    ptrue p0.d, vl1
+; CHECK-SD-NEXT:    mov z0.d, p0/m, x0
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: insert_vscale_2_i64_zero:
+; CHECK-GI:       // %bb.0: // %entry
+; CHECK-GI-NEXT:    mov x8, xzr
+; CHECK-GI-NEXT:    index z1.d, #0, #1
+; CHECK-GI-NEXT:    ptrue p0.d
+; CHECK-GI-NEXT:    mov z2.d, x8
+; CHECK-GI-NEXT:    cmpeq p0.d, p0/z, z1.d, z2.d
+; CHECK-GI-NEXT:    mov z0.d, p0/m, x0
+; CHECK-GI-NEXT:    ret
+entry:
+  %d = insertelement <vscale x 2 x i64> %vec, i64 %elt, i64 0
+  ret <vscale x 2 x i64> %d
+}
+
+define <vscale x 2 x i64> @insert_vscale_2_i64(<vscale x 2 x i64> %vec, i64 %elt, i64 %idx) {
+; CHECK-LABEL: insert_vscale_2_i64:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    index z1.d, #0, #1
+; CHECK-NEXT:    mov z2.d, x1
+; CHECK-NEXT:    ptrue p0.d
+; CHECK-NEXT:    cmpeq p0.d, p0/z, z1.d, z2.d
+; CHECK-NEXT:    mov z0.d, p0/m, x0
+; CHECK-NEXT:    ret
+entry:
+  %d = insertelement <vscale x  2 x i64> %vec, i64 %elt, i64 %idx
+  ret <vscale x 2 x i64> %d
+}
 
+define <vscale x 4 x i32> @insert_vscale_4_i32_zero(<vscale x 4 x i32> %vec, i32 %elt) {
+; CHECK-SD-LABEL: insert_vscale_4_i32_zero:
+; CHECK-SD:       // %bb.0: // %entry
+; CHECK-SD-NEXT:    ptrue p0.s, vl1
+; CHECK-SD-NEXT:    mov z0.s, p0/m, w0
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: insert_vscale_4_i32_zero:
+; CHECK-GI:       // %bb.0: // %entry
+; CHECK-GI-NEXT:    mov w8, wzr
+; CHECK-GI-NEXT:    index z1.s, #0, #1
+; CHECK-GI-NEXT:    ptrue p0.s
+; CHECK-GI-NEXT:    mov z2.s, w8
+; CHECK-GI-NEXT:    cmpeq p0.s, p0/z, z1.s, z2.s
+; CHECK-GI-NEXT:    mov z0.s, p0/m, w0
+; CHECK-GI-NEXT:    ret
+entry:
+  %d = insertelement <vscale x 4 x i32> %vec, i32 %elt, i64 0
+  ret <vscale x 4 x i32> %d
+}
+
+define <vscale x 4 x i32> @insert_vscale_4_i32(<vscale x 4 x i32> %vec, i32 %elt, i64 %idx) {
+; CHECK-LABEL: insert_vscale_4_i32:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    index z1.s, #0, #1
+; CHECK-NEXT:    mov z2.s, w1
+; CHECK-NEXT:    ptrue p0.s
+; CHECK-NEXT:    cmpeq p0.s, p0/z, z1.s, z2.s
+; CHECK-NEXT:    mov z0.s, p0/m, w0
+; CHECK-NEXT:    ret
+entry:
+  %d = insertelement <vscale x 4 x i32> %vec, i32 %elt, i64 %idx
+  ret <vscale x 4 x i32> %d
+}

@llvmbot
Copy link
Member

llvmbot commented Oct 31, 2024

@llvm/pr-subscribers-llvm-globalisel

Author: Thorsten Schütt (tschuett)

Changes

There are patterns for:

  • {nxv2s32, s32, s64},
  • {nxv4s16, s16, s64},
  • {nxv2s16, s16, s64}

For {nxv16s8, s32, s64}, {nxv8s16, s32, s64} element type and vector element type differ, which makes LLVM-IR unhappy.

Full diff: https://github.com/llvm/llvm-project/pull/114470.diff

5 Files Affected:

  • (modified) llvm/include/llvm/CodeGen/GlobalISel/LegalizerInfo.h (+20)
  • (modified) llvm/lib/CodeGen/GlobalISel/LegalityPredicates.cpp (+11)
  • (modified) llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp (+1)
  • (modified) llvm/lib/Target/AArch64/GISel/AArch64PostLegalizerLowering.cpp (+3)
  • (modified) llvm/test/CodeGen/AArch64/extract-vector-elt.ll (+76-8)
diff --git a/llvm/include/llvm/CodeGen/GlobalISel/LegalizerInfo.h b/llvm/include/llvm/CodeGen/GlobalISel/LegalizerInfo.h
index 6d71c150c8da6b..343a0172ff39ed 100644
--- a/llvm/include/llvm/CodeGen/GlobalISel/LegalizerInfo.h
+++ b/llvm/include/llvm/CodeGen/GlobalISel/LegalizerInfo.h
@@ -273,6 +273,11 @@ inline LegalityPredicate typeIsNot(unsigned TypeIdx, LLT Type) {
 LegalityPredicate
 typePairInSet(unsigned TypeIdx0, unsigned TypeIdx1,
               std::initializer_list<std::pair<LLT, LLT>> TypesInit);
+/// True iff the given types for the given tuple of type indexes is one of the
+/// specified type tuple.
+LegalityPredicate
+typeTupleInSet(unsigned TypeIdx0, unsigned TypeIdx1, unsigned Type2,
+               std::initializer_list<std::tuple<LLT, LLT, LLT>> TypesInit);
 /// True iff the given types for the given pair of type indexes is one of the
 /// specified type pairs.
 LegalityPredicate typePairAndMemDescInSet(
@@ -504,6 +509,15 @@ class LegalizeRuleSet {
     using namespace LegalityPredicates;
     return actionIf(Action, typePairInSet(typeIdx(0), typeIdx(1), Types));
   }
+
+  LegalizeRuleSet &
+  actionFor(LegalizeAction Action,
+            std::initializer_list<std::tuple<LLT, LLT, LLT>> Types) {
+    using namespace LegalityPredicates;
+    return actionIf(Action,
+                    typeTupleInSet(typeIdx(0), typeIdx(1), typeIdx(2), Types));
+  }
+
   /// Use the given action when type indexes 0 and 1 is any type pair in the
   /// given list.
   /// Action should be an action that requires mutation.
@@ -615,6 +629,12 @@ class LegalizeRuleSet {
       return *this;
     return actionFor(LegalizeAction::Legal, Types);
   }
+  LegalizeRuleSet &
+  legalFor(bool Pred, std::initializer_list<std::tuple<LLT, LLT, LLT>> Types) {
+    if (!Pred)
+      return *this;
+    return actionFor(LegalizeAction::Legal, Types);
+  }
   /// The instruction is legal when type index 0 is any type in the given list
   /// and imm index 0 is anything.
   LegalizeRuleSet &legalForTypeWithAnyImm(std::initializer_list<LLT> Types) {
diff --git a/llvm/lib/CodeGen/GlobalISel/LegalityPredicates.cpp b/llvm/lib/CodeGen/GlobalISel/LegalityPredicates.cpp
index 8fe48195c610be..dc7ed6cbe8b7da 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegalityPredicates.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegalityPredicates.cpp
@@ -49,6 +49,17 @@ LegalityPredicate LegalityPredicates::typePairInSet(
   };
 }
 
+LegalityPredicate LegalityPredicates::typeTupleInSet(
+    unsigned TypeIdx0, unsigned TypeIdx1, unsigned TypeIdx2,
+    std::initializer_list<std::tuple<LLT, LLT, LLT>> TypesInit) {
+  SmallVector<std::tuple<LLT, LLT, LLT>, 4> Types = TypesInit;
+  return [=](const LegalityQuery &Query) {
+    std::tuple<LLT, LLT, LLT> Match = {
+        Query.Types[TypeIdx0], Query.Types[TypeIdx1], Query.Types[TypeIdx2]};
+    return llvm::is_contained(Types, Match);
+  };
+}
+
 LegalityPredicate LegalityPredicates::typePairAndMemDescInSet(
     unsigned TypeIdx0, unsigned TypeIdx1, unsigned MMOIdx,
     std::initializer_list<TypePairAndMemDesc> TypesAndMemDescInit) {
diff --git a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
index 400024922124cd..31627060c9a6bc 100644
--- a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
+++ b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
@@ -978,6 +978,7 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
   getActionDefinitionsBuilder(G_INSERT_VECTOR_ELT)
       .legalIf(
           typeInSet(0, {v16s8, v8s8, v8s16, v4s16, v4s32, v2s32, v2s64, v2p0}))
+      .legalFor(HasSVE, {{nxv4s32, s32, s64}, {nxv2s64, s64, s64}})
       .moreElementsToNextPow2(0)
       .widenVectorEltsToVectorMinSize(0, 64)
       .clampNumElements(0, v8s8, v16s8)
diff --git a/llvm/lib/Target/AArch64/GISel/AArch64PostLegalizerLowering.cpp b/llvm/lib/Target/AArch64/GISel/AArch64PostLegalizerLowering.cpp
index b40fe55fdfaf67..2bcfdc1b46873b 100644
--- a/llvm/lib/Target/AArch64/GISel/AArch64PostLegalizerLowering.cpp
+++ b/llvm/lib/Target/AArch64/GISel/AArch64PostLegalizerLowering.cpp
@@ -426,6 +426,9 @@ void applyNonConstInsert(MachineInstr &MI, MachineRegisterInfo &MRI,
   LLT EltTy = MRI.getType(Insert.getElementReg());
   LLT IdxTy = MRI.getType(Insert.getIndexReg());
 
+  if (VecTy.isScalableVector())
+    return;
+
   // Create a stack slot and store the vector into it
   MachineFunction &MF = Builder.getMF();
   Align Alignment(
diff --git a/llvm/test/CodeGen/AArch64/extract-vector-elt.ll b/llvm/test/CodeGen/AArch64/extract-vector-elt.ll
index 0481d997d24faf..b7cc04fa6edc58 100644
--- a/llvm/test/CodeGen/AArch64/extract-vector-elt.ll
+++ b/llvm/test/CodeGen/AArch64/extract-vector-elt.ll
@@ -1,6 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
-; RUN: llc -mtriple=aarch64 -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-SD
-; RUN: llc -mtriple=aarch64 -global-isel -global-isel-abort=2 -verify-machineinstrs %s -o - 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-GI
+; RUN: llc -mtriple=aarch64 -mattr=+sve -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-SD
+; RUN: llc -mtriple=aarch64 -mattr=+sve -aarch64-enable-gisel-sve=1 -global-isel -global-isel-abort=2 -verify-machineinstrs %s -o - 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-GI
 
 ; CHECK-GI:       warning: Instruction selection used fallback path for extract_v4i32_vector_insert
 ; CHECK-GI-NEXT:  warning: Instruction selection used fallback path for extract_v4i32_vector_insert_const
@@ -143,11 +143,12 @@ define i64 @extract_v2i64_extract_build_vector_opaque(<2 x i64> %a, i32 %c) {
 ; CHECK-SD:       // %bb.0: // %entry
 ; CHECK-SD-NEXT:    sub sp, sp, #16
 ; CHECK-SD-NEXT:    .cfi_def_cfa_offset 16
-; CHECK-SD-NEXT:    adrp x8, .LCPI8_0
+; CHECK-SD-NEXT:    mov x8, #-31 // =0xffffffffffffffe1
 ; CHECK-SD-NEXT:    // kill: def $w0 killed $w0 def $x0
-; CHECK-SD-NEXT:    ldr q0, [x8, :lo12:.LCPI8_0]
+; CHECK-SD-NEXT:    index z0.d, #0, x8
 ; CHECK-SD-NEXT:    mov x8, sp
 ; CHECK-SD-NEXT:    bfi x8, x0, #3, #1
+; CHECK-SD-NEXT:    add z0.d, z0.d, #42 // =0x2a
 ; CHECK-SD-NEXT:    str q0, [sp]
 ; CHECK-SD-NEXT:    ldr x0, [x8]
 ; CHECK-SD-NEXT:    add sp, sp, #16
@@ -1043,11 +1044,10 @@ entry:
 define i32 @extract_v4i32_phi(i64 %val, i32  %limit, ptr %ptr) {
 ; CHECK-SD-LABEL: extract_v4i32_phi:
 ; CHECK-SD:       // %bb.0: // %entry
-; CHECK-SD-NEXT:    dup v1.2s, w0
-; CHECK-SD-NEXT:    adrp x8, .LCPI41_0
+; CHECK-SD-NEXT:    index z1.s, #1, #1
+; CHECK-SD-NEXT:    dup v2.2s, w0
 ; CHECK-SD-NEXT:    movi v0.2s, #16
-; CHECK-SD-NEXT:    ldr d2, [x8, :lo12:.LCPI41_0]
-; CHECK-SD-NEXT:    add v1.2s, v1.2s, v2.2s
+; CHECK-SD-NEXT:    add v1.2s, v2.2s, v1.2s
 ; CHECK-SD-NEXT:  .LBB41_1: // %loop
 ; CHECK-SD-NEXT:    // =>This Inner Loop Header: Depth=1
 ; CHECK-SD-NEXT:    fmov w8, s1
@@ -1100,4 +1100,72 @@ ret:
   ret i32 %3
 }
 
+define <vscale x 2 x i64> @insert_vscale_2_i64_zero(<vscale x 2 x i64> %vec, i64 %elt) {
+; CHECK-SD-LABEL: insert_vscale_2_i64_zero:
+; CHECK-SD:       // %bb.0: // %entry
+; CHECK-SD-NEXT:    ptrue p0.d, vl1
+; CHECK-SD-NEXT:    mov z0.d, p0/m, x0
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: insert_vscale_2_i64_zero:
+; CHECK-GI:       // %bb.0: // %entry
+; CHECK-GI-NEXT:    mov x8, xzr
+; CHECK-GI-NEXT:    index z1.d, #0, #1
+; CHECK-GI-NEXT:    ptrue p0.d
+; CHECK-GI-NEXT:    mov z2.d, x8
+; CHECK-GI-NEXT:    cmpeq p0.d, p0/z, z1.d, z2.d
+; CHECK-GI-NEXT:    mov z0.d, p0/m, x0
+; CHECK-GI-NEXT:    ret
+entry:
+  %d = insertelement <vscale x 2 x i64> %vec, i64 %elt, i64 0
+  ret <vscale x 2 x i64> %d
+}
+
+define <vscale x 2 x i64> @insert_vscale_2_i64(<vscale x 2 x i64> %vec, i64 %elt, i64 %idx) {
+; CHECK-LABEL: insert_vscale_2_i64:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    index z1.d, #0, #1
+; CHECK-NEXT:    mov z2.d, x1
+; CHECK-NEXT:    ptrue p0.d
+; CHECK-NEXT:    cmpeq p0.d, p0/z, z1.d, z2.d
+; CHECK-NEXT:    mov z0.d, p0/m, x0
+; CHECK-NEXT:    ret
+entry:
+  %d = insertelement <vscale x  2 x i64> %vec, i64 %elt, i64 %idx
+  ret <vscale x 2 x i64> %d
+}
 
+define <vscale x 4 x i32> @insert_vscale_4_i32_zero(<vscale x 4 x i32> %vec, i32 %elt) {
+; CHECK-SD-LABEL: insert_vscale_4_i32_zero:
+; CHECK-SD:       // %bb.0: // %entry
+; CHECK-SD-NEXT:    ptrue p0.s, vl1
+; CHECK-SD-NEXT:    mov z0.s, p0/m, w0
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: insert_vscale_4_i32_zero:
+; CHECK-GI:       // %bb.0: // %entry
+; CHECK-GI-NEXT:    mov w8, wzr
+; CHECK-GI-NEXT:    index z1.s, #0, #1
+; CHECK-GI-NEXT:    ptrue p0.s
+; CHECK-GI-NEXT:    mov z2.s, w8
+; CHECK-GI-NEXT:    cmpeq p0.s, p0/z, z1.s, z2.s
+; CHECK-GI-NEXT:    mov z0.s, p0/m, w0
+; CHECK-GI-NEXT:    ret
+entry:
+  %d = insertelement <vscale x 4 x i32> %vec, i32 %elt, i64 0
+  ret <vscale x 4 x i32> %d
+}
+
+define <vscale x 4 x i32> @insert_vscale_4_i32(<vscale x 4 x i32> %vec, i32 %elt, i64 %idx) {
+; CHECK-LABEL: insert_vscale_4_i32:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    index z1.s, #0, #1
+; CHECK-NEXT:    mov z2.s, w1
+; CHECK-NEXT:    ptrue p0.s
+; CHECK-NEXT:    cmpeq p0.s, p0/z, z1.s, z2.s
+; CHECK-NEXT:    mov z0.s, p0/m, w0
+; CHECK-NEXT:    ret
+entry:
+  %d = insertelement <vscale x 4 x i32> %vec, i32 %elt, i64 %idx
+  ret <vscale x 4 x i32> %d
+}

@tschuett
Copy link
Author

Last PR had only MIR tests, which hid the LLVM-IR issue and made Windows CI unhappy.

arsenm
arsenm reviewed Oct 31, 2024
View reviewed changes
arsenm
arsenm approved these changes Oct 31, 2024
View reviewed changes
Copy link
Contributor

@arsenm arsenm left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

lgtm with nit

llvm/test/CodeGen/AArch64/extract-vector-elt.ll
@@ -1099,5 +1099,3 @@ loop:
ret:
ret i32 %3
}

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Leftover whitespace change

Copy link
Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Trunk has no whitespace:
https://github.com/llvm/llvm-project/blob/main/llvm/test/CodeGen/AArch64/extract-vector-elt.ll

@tschuett tschuett merged commit 8e37727 into llvm:main Nov 1, 2024
6 of 8 checks passed
@tschuett tschuett deleted the gisel-legalize-insert-vector-elt2 branch November 1, 2024 05:10
smallp-o-p pushed a commit to smallp-o-p/llvm-project that referenced this pull request Nov 3, 2024
@tschuett
[GlobalISel][AArch64] Legalize G_INSERT_VECTOR_ELT for SVE (llvm#114470)
5ee9571 
There are patterns for:
* {nxv2s32, s32, s64},
* {nxv4s16, s16, s64},
* {nxv2s16, s16, s64}
NoumanAmir657 pushed a commit to NoumanAmir657/llvm-project that referenced this pull request Nov 4, 2024
@tschuett @NoumanAmir657
[GlobalISel][AArch64] Legalize G_INSERT_VECTOR_ELT for SVE (llvm#114470)
c0c30d7 
There are patterns for:
* {nxv2s32, s32, s64},
* {nxv4s16, s16, s64},
* {nxv2s16, s16, s64}
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@arsenm arsenm arsenm approved these changes

@davemgreen davemgreen Awaiting requested review from davemgreen

Assignees
No one assigned
Labels
backend:AArch64 llvm:globalisel
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@tschuett @llvmbot @arsenm