[PATCH] D135821: AMDGPU: Fix assertion on <1 x i16> vectors

[Matt Arsenault via Phabricator via llvm-commits]

arsenm created this revision.
arsenm added a reviewer: AMDGPU.
Herald added subscribers: kosarev, foad, kerbowa, hiraditya, t-tye, tpr, dstuttard, yaxunl, jvesely, kzhuravl.
Herald added a project: All.
arsenm requested review of this revision.
Herald added a subscriber: wdng.
Herald added a project: LLVM.

Fixes issue 58331.


https://reviews.llvm.org/D135821

Files:
  llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp
  llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-mul24.ll


Index: llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-mul24.ll
===================================================================
--- llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-mul24.ll
+++ llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-mul24.ll
@@ -56,6 +56,35 @@
   ret i32 %mul
 }
 
+define <2 x i8> @f(<1 x i16> %arg) {
+; SI-LABEL: @f(
+; SI-NEXT:  BB:
+; SI-NEXT:    [[TMP0:%.*]] = extractelement <1 x i16> [[ARG:%.*]], i64 0
+; SI-NEXT:    [[TMP1:%.*]] = zext i16 [[TMP0]] to i32
+; SI-NEXT:    [[TMP2:%.*]] = call i32 @llvm.amdgcn.mul.u24(i32 [[TMP1]], i32 42)
+; SI-NEXT:    [[TMP3:%.*]] = trunc i32 [[TMP2]] to i16
+; SI-NEXT:    [[MUL:%.*]] = insertelement <1 x i16> undef, i16 [[TMP3]], i64 0
+; SI-NEXT:    [[CAST:%.*]] = bitcast <1 x i16> [[MUL]] to <2 x i8>
+; SI-NEXT:    ret <2 x i8> [[CAST]]
+;
+; VI-LABEL: @f(
+; VI-NEXT:  BB:
+; VI-NEXT:    [[MUL:%.*]] = mul <1 x i16> [[ARG:%.*]], <i16 42>
+; VI-NEXT:    [[CAST:%.*]] = bitcast <1 x i16> [[MUL]] to <2 x i8>
+; VI-NEXT:    ret <2 x i8> [[CAST]]
+;
+; DISABLED-LABEL: @f(
+; DISABLED-NEXT:  BB:
+; DISABLED-NEXT:    [[MUL:%.*]] = mul <1 x i16> [[ARG:%.*]], <i16 42>
+; DISABLED-NEXT:    [[CAST:%.*]] = bitcast <1 x i16> [[MUL]] to <2 x i8>
+; DISABLED-NEXT:    ret <2 x i8> [[CAST]]
+;
+BB:
+  %mul = mul <1 x i16> %arg, <i16 42>
+  %cast = bitcast <1 x i16> %mul to <2 x i8>
+  ret <2 x i8> %cast
+}
+
 define <2 x i32> @smul24_v2i32(<2 x i32> %lhs, <2 x i32> %rhs) {
 ; SI-LABEL: @smul24_v2i32(
 ; SI-NEXT:    [[SHL_LHS:%.*]] = shl <2 x i32> [[LHS:%.*]], <i32 8, i32 8>
Index: llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp
===================================================================
--- llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp
+++ llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp
@@ -468,8 +468,10 @@
 static Value *insertValues(IRBuilder<> &Builder,
                            Type *Ty,
                            SmallVectorImpl<Value *> &Values) {
-  if (Values.size() == 1)
+  if (!Ty->isVectorTy()) {
+    assert(Values.size() == 1);
     return Values[0];
+  }
 
   Value *NewVal = UndefValue::get(Ty);
   for (int I = 0, E = Values.size(); I != E; ++I)


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D135821.467278.patch
Type: text/x-patch
Size: 2157 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20221012/c7ad2a02/attachment.bin>