[llvm] eaf23b2 - [GIsel][AArch64] Legalize <2 x i16> for G_INSERT_VECTOR_ELT (#65830)
via llvm-commits
llvm-commits at lists.llvm.org
Tue Sep 12 06:15:06 PDT 2023
Author: Allen
Date: 2023-09-12T21:15:01+08:00
New Revision: eaf23b2480a1c892d86abb7c911d86687a7cfe72
URL: https://github.com/llvm/llvm-project/commit/eaf23b2480a1c892d86abb7c911d86687a7cfe72
DIFF: https://github.com/llvm/llvm-project/commit/eaf23b2480a1c892d86abb7c911d86687a7cfe72.diff
LOG: [GIsel][AArch64] Legalize <2 x i16> for G_INSERT_VECTOR_ELT (#65830)
Widen the vector elements to 64 bits to make sure it legal instead by
clamping the number of elements. Depend on D153394.
Fixes https://github.com/llvm/llvm-project/issues/63826
Added:
llvm/test/CodeGen/AArch64/GlobalISel/insert-vector-elt-pr63826.ll
Modified:
llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
llvm/test/CodeGen/AArch64/GlobalISel/legalize-insert-vector-elt.mir
Removed:
################################################################################
diff --git a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
index cfb95955d1f888b..a5e8024090866dc 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
@@ -2496,6 +2496,17 @@ LegalizerHelper::widenScalar(MachineInstr &MI, unsigned TypeIdx, LLT WideTy) {
return Legalized;
}
case TargetOpcode::G_INSERT_VECTOR_ELT: {
+ if (TypeIdx == 0) {
+ Observer.changingInstr(MI);
+ const LLT WideEltTy = WideTy.getElementType();
+
+ widenScalarSrc(MI, WideTy, 1, TargetOpcode::G_ANYEXT);
+ widenScalarSrc(MI, WideEltTy, 2, TargetOpcode::G_ANYEXT);
+ widenScalarDst(MI, WideTy, 0);
+ Observer.changedInstr(MI);
+ return Legalized;
+ }
+
if (TypeIdx == 1) {
Observer.changingInstr(MI);
diff --git a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
index a8065be726e7da2..9ea4e349e69b997 100644
--- a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
+++ b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
@@ -732,8 +732,7 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
getActionDefinitionsBuilder(G_INSERT_VECTOR_ELT)
.legalIf(typeInSet(0, {v16s8, v8s8, v8s16, v4s16, v4s32, v2s32, v2s64}))
- .clampMinNumElements(0, s16, 4)
- .clampMaxNumElements(0, s16, 8);
+ .widenVectorEltsToVectorMinSize(0, 64);
getActionDefinitionsBuilder(G_BUILD_VECTOR)
.legalFor({{v8s8, s8},
diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/insert-vector-elt-pr63826.ll b/llvm/test/CodeGen/AArch64/GlobalISel/insert-vector-elt-pr63826.ll
new file mode 100644
index 000000000000000..e41b25fb2b88f14
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/insert-vector-elt-pr63826.ll
@@ -0,0 +1,42 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -o - -verify-machineinstrs -global-isel=1 -global-isel-abort=1 | FileCheck %s --check-prefixes=CHECK
+; RUN: llc < %s -o - -verify-machineinstrs -global-isel=0 | FileCheck %s --check-prefixes=CHECK
+
+target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"
+target triple = "arm64-apple-ios14.5.0"
+
+define <2 x i16> @pr63826_v2s16(<2 x i16> %vec) {
+; CHECK-LABEL: pr63826_v2s16:
+; CHECK: ; %bb.0:
+; CHECK-NEXT: ; kill: def $d0 killed $d0 def $q0
+; CHECK-NEXT: mov w8, #1 ; =0x1
+; CHECK-NEXT: mov.s v0[0], w8
+; CHECK-NEXT: ; kill: def $d0 killed $d0 killed $q0
+; CHECK-NEXT: ret
+ %vec1 = insertelement <2 x i16> %vec, i16 1, i32 0
+ ret <2 x i16> %vec1
+}
+
+define <2 x i8> @pr63826_v2s8(<2 x i8> %vec) {
+; CHECK-LABEL: pr63826_v2s8:
+; CHECK: ; %bb.0:
+; CHECK-NEXT: ; kill: def $d0 killed $d0 def $q0
+; CHECK-NEXT: mov w8, #1 ; =0x1
+; CHECK-NEXT: mov.s v0[0], w8
+; CHECK-NEXT: ; kill: def $d0 killed $d0 killed $q0
+; CHECK-NEXT: ret
+ %vec1 = insertelement <2 x i8> %vec, i8 1, i32 0
+ ret <2 x i8> %vec1
+}
+
+define <4 x i8> @pr63826_v4s8(<4 x i8> %vec) {
+; CHECK-LABEL: pr63826_v4s8:
+; CHECK: ; %bb.0:
+; CHECK-NEXT: ; kill: def $d0 killed $d0 def $q0
+; CHECK-NEXT: mov w8, #1 ; =0x1
+; CHECK-NEXT: mov.h v0[0], w8
+; CHECK-NEXT: ; kill: def $d0 killed $d0 killed $q0
+; CHECK-NEXT: ret
+ %vec1 = insertelement <4 x i8> %vec, i8 1, i32 0
+ ret <4 x i8> %vec1
+}
diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/legalize-insert-vector-elt.mir b/llvm/test/CodeGen/AArch64/GlobalISel/legalize-insert-vector-elt.mir
index 26db18bd611a57c..6f6cf2cc165b9f1 100644
--- a/llvm/test/CodeGen/AArch64/GlobalISel/legalize-insert-vector-elt.mir
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/legalize-insert-vector-elt.mir
@@ -1,24 +1,73 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=aarch64-linux-gnu -O0 -run-pass=legalizer %s -o - -global-isel-abort=1 | FileCheck %s
---
-name: pr63826
+name: pr63826_v2s16
body: |
bb.0:
- ; CHECK-LABEL: name: pr63826
- ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $w0
- ; CHECK-NEXT: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 1
- ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 42
- ; CHECK-NEXT: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
- ; CHECK-NEXT: [[DEF:%[0-9]+]]:_(s16) = G_IMPLICIT_DEF
- ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[UV]](s16), [[UV1]](s16), [[DEF]](s16), [[DEF]](s16)
- ; CHECK-NEXT: [[IVEC:%[0-9]+]]:_(<4 x s16>) = G_INSERT_VECTOR_ELT [[BUILD_VECTOR]], [[C]](s16), [[C1]](s32)
- ; CHECK-NEXT: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[IVEC]](<4 x s16>)
- ; CHECK-NEXT: $w0 = COPY [[UV2]](<2 x s16>)
- %0:_(<2 x s16>) = COPY $w0
- %1:_(s16) = G_CONSTANT i16 1
- %2:_(s32) = G_CONSTANT i32 42
- %4:_(<2 x s16>) = G_INSERT_VECTOR_ELT %0(<2 x s16>), %1(s16), %2(s32)
- $w0 = COPY %4(<2 x s16>)
+ liveins: $d0
+ ; CHECK-LABEL: name: pr63826_v2s16
+ ; CHECK: liveins: $d0
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $d0
+ ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+ ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
+ ; CHECK-NEXT: [[IVEC:%[0-9]+]]:_(<2 x s32>) = G_INSERT_VECTOR_ELT [[COPY]], [[C1]](s32), [[C]](s32)
+ ; CHECK-NEXT: $d0 = COPY [[IVEC]](<2 x s32>)
+ ; CHECK-NEXT: RET_ReallyLR implicit $d0
+ %1:_(<2 x s32>) = COPY $d0
+ %0:_(<2 x s16>) = G_TRUNC %1(<2 x s32>)
+ %4:_(s32) = G_CONSTANT i32 0
+ %3:_(s16) = G_CONSTANT i16 1
+ %2:_(<2 x s16>) = G_INSERT_VECTOR_ELT %0, %3(s16), %4(s32)
+ %5:_(<2 x s32>) = G_ANYEXT %2(<2 x s16>)
+ $d0 = COPY %5(<2 x s32>)
+ RET_ReallyLR implicit $d0
+...
+---
+name: pr63826_v2s8
+body: |
+ bb.0:
+ liveins: $d0
+ ; CHECK-LABEL: name: pr63826_v2s8
+ ; CHECK: liveins: $d0
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $d0
+ ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+ ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
+ ; CHECK-NEXT: [[IVEC:%[0-9]+]]:_(<2 x s32>) = G_INSERT_VECTOR_ELT [[COPY]], [[C1]](s32), [[C]](s32)
+ ; CHECK-NEXT: $d0 = COPY [[IVEC]](<2 x s32>)
+ ; CHECK-NEXT: RET_ReallyLR implicit $d0
+ %1:_(<2 x s32>) = COPY $d0
+ %0:_(<2 x s8>) = G_TRUNC %1(<2 x s32>)
+ %4:_(s32) = G_CONSTANT i32 0
+ %3:_(s8) = G_CONSTANT i8 1
+ %2:_(<2 x s8>) = G_INSERT_VECTOR_ELT %0, %3(s8), %4(s32)
+ %5:_(<2 x s32>) = G_ANYEXT %2(<2 x s8>)
+ $d0 = COPY %5(<2 x s32>)
+ RET_ReallyLR implicit $d0
+...
+---
+name: pr63826_v4s8
+body: |
+ bb.0:
+ liveins: $d0
+ ; CHECK-LABEL: name: pr63826_v4s8
+ ; CHECK: liveins: $d0
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $d0
+ ; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
+ ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 1
+ ; CHECK-NEXT: [[IVEC:%[0-9]+]]:_(<4 x s16>) = G_INSERT_VECTOR_ELT [[COPY]], [[C1]](s16), [[C]](s32)
+ ; CHECK-NEXT: $d0 = COPY [[IVEC]](<4 x s16>)
+ ; CHECK-NEXT: RET_ReallyLR implicit $d0
+ %1:_(<4 x s16>) = COPY $d0
+ %0:_(<4 x s8>) = G_TRUNC %1(<4 x s16>)
+ %4:_(s32) = G_CONSTANT i32 0
+ %3:_(s8) = G_CONSTANT i8 1
+ %2:_(<4 x s8>) = G_INSERT_VECTOR_ELT %0, %3(s8), %4(s32)
+ %5:_(<4 x s16>) = G_ANYEXT %2(<4 x s8>)
+ $d0 = COPY %5(<4 x s16>)
+ RET_ReallyLR implicit $d0
...
---
name: v8s8
More information about the llvm-commits
mailing list