[PATCH] D65413: GlobalISel: Implement moreElementsVector for G_UNMERGE_VALUES sources

[Matt Arsenault via Phabricator via llvm-commits]

arsenm updated this revision to Diff 212217.
arsenm added a comment.

Attach right version


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D65413/new/

https://reviews.llvm.org/D65413

Files:
  lib/CodeGen/GlobalISel/LegalizerHelper.cpp
  lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
  test/CodeGen/AMDGPU/GlobalISel/legalize-fexp.mir


Index: test/CodeGen/AMDGPU/GlobalISel/legalize-fexp.mir
===================================================================
--- test/CodeGen/AMDGPU/GlobalISel/legalize-fexp.mir
+++ test/CodeGen/AMDGPU/GlobalISel/legalize-fexp.mir
@@ -9,8 +9,8 @@
 
     ; CHECK-LABEL: name: test_fexp_s32
     ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
-    ; CHECK: [[FEXP_:%[0-9]+]]:_(s32) = G_FEXP [[COPY]]
-    ; CHECK: $vgpr0 = COPY [[FEXP_]](s32)
+    ; CHECK: [[FEXP:%[0-9]+]]:_(s32) = G_FEXP [[COPY]]
+    ; CHECK: $vgpr0 = COPY [[FEXP]](s32)
     %0:_(s32) = COPY $vgpr0
     %1:_(s32) = G_FEXP %0
     $vgpr0 = COPY %1
@@ -25,9 +25,9 @@
     ; CHECK-LABEL: name: test_fexp_v2s32
     ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
     ; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
-    ; CHECK: [[FEXP_:%[0-9]+]]:_(s32) = G_FEXP [[UV]]
-    ; CHECK: [[FEXP_1:%[0-9]+]]:_(s32) = G_FEXP [[UV1]]
-    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FEXP_]](s32), [[FEXP_1]](s32)
+    ; CHECK: [[FEXP:%[0-9]+]]:_(s32) = G_FEXP [[UV]]
+    ; CHECK: [[FEXP1:%[0-9]+]]:_(s32) = G_FEXP [[UV1]]
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FEXP]](s32), [[FEXP1]](s32)
     ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
     %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
     %1:_(<2 x s32>) = G_FEXP %0
@@ -43,10 +43,10 @@
     ; CHECK-LABEL: name: test_fexp_v3s32
     ; CHECK: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
     ; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
-    ; CHECK: [[FEXP_:%[0-9]+]]:_(s32) = G_FEXP [[UV]]
-    ; CHECK: [[FEXP_1:%[0-9]+]]:_(s32) = G_FEXP [[UV1]]
-    ; CHECK: [[FEXP_2:%[0-9]+]]:_(s32) = G_FEXP [[UV2]]
-    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FEXP_]](s32), [[FEXP_1]](s32), [[FEXP_2]](s32)
+    ; CHECK: [[FEXP:%[0-9]+]]:_(s32) = G_FEXP [[UV]]
+    ; CHECK: [[FEXP1:%[0-9]+]]:_(s32) = G_FEXP [[UV1]]
+    ; CHECK: [[FEXP2:%[0-9]+]]:_(s32) = G_FEXP [[UV2]]
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FEXP]](s32), [[FEXP1]](s32), [[FEXP2]](s32)
     ; CHECK: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
     %0:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
     %1:_(<3 x  s32>) = G_FEXP %0
Index: lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
===================================================================
--- lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -929,6 +929,7 @@
                            elementTypeIs(1, S16)),
                        changeTo(1, V2S16))
       // Break up vectors with weird elements into scalars
+      .moreElementsIf(isSmallOddVector(BigTyIdx), oneMoreElement(BigTyIdx))
       .fewerElementsIf(
         [=](const LegalityQuery &Query) { return notValidElt(Query, 0); },
         scalarize(0))
Index: lib/CodeGen/GlobalISel/LegalizerHelper.cpp
===================================================================
--- lib/CodeGen/GlobalISel/LegalizerHelper.cpp
+++ lib/CodeGen/GlobalISel/LegalizerHelper.cpp
@@ -2875,6 +2875,26 @@
     moreElementsVectorDst(MI, MoreTy, 0);
     Observer.changedInstr(MI);
     return Legalized;
+  case TargetOpcode::G_UNMERGE_VALUES: {
+    if (TypeIdx != 1)
+      return UnableToLegalize;
+
+    LLT DstTy = MRI.getType(MI.getOperand(0).getReg());
+    int NumDst = MI.getNumOperands() - 1;
+    moreElementsVectorSrc(MI, MoreTy, NumDst);
+
+    auto MIB = MIRBuilder.buildInstr(TargetOpcode::G_UNMERGE_VALUES);
+    for (int I = 0; I != NumDst; ++I)
+      MIB.addDef(MI.getOperand(I).getReg());
+
+    int NewNumDst = MoreTy.getSizeInBits() / DstTy.getSizeInBits();
+    for (int I = NumDst; I != NewNumDst; ++I)
+      MIB.addDef(MRI.createGenericVirtualRegister(DstTy));
+
+    MIB.addUse(MI.getOperand(NumDst).getReg());
+    MI.eraseFromParent();
+    return Legalized;
+  }
   case TargetOpcode::G_PHI:
     return moreElementsVectorPhi(MI, TypeIdx, MoreTy);
   default:


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D65413.212217.patch
Type: text/x-patch
Size: 4074 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20190729/297d785a/attachment.bin>