From 1b59f4c380659e330ff3c03ddcfd0d6547dfd43a Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault@amd.com>
Date: Wed, 20 Feb 2019 16:42:52 +0000
Subject: [PATCH] GlobalISel: Fix fewerElementsVector for ctlz with different
 result type

Also complete the set of related operations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@354480 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/CodeGen/GlobalISel/LegalizerHelper.cpp    |  6 +-
 lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp     |  4 +-
 .../GlobalISel/legalize-ctlz-zero-undef.mir   | 64 +++++++++++++++++++
 .../AMDGPU/GlobalISel/legalize-ctlz.mir       | 64 +++++++++++++++++++
 .../AMDGPU/GlobalISel/legalize-ctpop.mir      | 60 +++++++++++++++++
 .../GlobalISel/legalize-cttz-zero-undef.mir   | 60 +++++++++++++++++
 .../AMDGPU/GlobalISel/legalize-cttz.mir       | 64 +++++++++++++++++++
 7 files changed, 319 insertions(+), 3 deletions(-)

diff --git a/lib/CodeGen/GlobalISel/LegalizerHelper.cpp b/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
index 82ba0b8d09c9..7abce5a3fa89 100644
--- a/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
+++ b/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
@@ -2129,11 +2129,15 @@ LegalizerHelper::fewerElementsVector(MachineInstr &MI, unsigned TypeIdx,
   case G_FSIN:
   case G_FSQRT:
   case G_BSWAP:
-  case G_CTLZ:
     return fewerElementsVectorBasic(MI, TypeIdx, NarrowTy);
   case G_SHL:
   case G_LSHR:
   case G_ASHR:
+  case G_CTLZ:
+  case G_CTLZ_ZERO_UNDEF:
+  case G_CTTZ:
+  case G_CTTZ_ZERO_UNDEF:
+  case G_CTPOP:
     return fewerElementsVectorMultiEltType(MI, TypeIdx, NarrowTy);
   case G_ZEXT:
   case G_SEXT:
diff --git a/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
index 300f5c18870c..0d0496eaa035 100644
--- a/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ b/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -306,8 +306,8 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST,
                                G_CTPOP})
     .legalFor({{S32, S32}, {S32, S64}})
     .clampScalar(0, S32, S32)
-    .clampScalar(1, S32, S64);
-  // TODO: Scalarize
+    .clampScalar(1, S32, S64)
+    .scalarize(0);
 
   // TODO: Expand for > s32
   getActionDefinitionsBuilder(G_BSWAP)
diff --git a/test/CodeGen/AMDGPU/GlobalISel/legalize-ctlz-zero-undef.mir b/test/CodeGen/AMDGPU/GlobalISel/legalize-ctlz-zero-undef.mir
index aa01879f7bf6..ff93788bc86b 100644
--- a/test/CodeGen/AMDGPU/GlobalISel/legalize-ctlz-zero-undef.mir
+++ b/test/CodeGen/AMDGPU/GlobalISel/legalize-ctlz-zero-undef.mir
@@ -91,3 +91,67 @@ body: |
     %3:_(s32) = G_ZEXT %2
     $vgpr0 = COPY %3
 ...
+
+---
+name: ctlz_zero_undef_v2s32_v2s32
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+    ; CHECK-LABEL: name: ctlz_zero_undef_v2s32_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: [[CTLZ_ZERO_UNDEF:%[0-9]+]]:_(s32) = G_CTLZ_ZERO_UNDEF [[UV]](s32)
+    ; CHECK: [[CTLZ_ZERO_UNDEF1:%[0-9]+]]:_(s32) = G_CTLZ_ZERO_UNDEF [[UV1]](s32)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[CTLZ_ZERO_UNDEF]](s32), [[CTLZ_ZERO_UNDEF1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = G_CTLZ_ZERO_UNDEF %0
+    $vgpr0_vgpr1 = COPY %1
+...
+
+---
+name: ctlz_zero_undef_v2s32_v2s64
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2_vgpr3
+    ; CHECK-LABEL: name: ctlz_zero_undef_v2s32_v2s64
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; CHECK: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; CHECK: [[CTLZ_ZERO_UNDEF:%[0-9]+]]:_(s32) = G_CTLZ_ZERO_UNDEF [[UV]](s64)
+    ; CHECK: [[CTLZ_ZERO_UNDEF1:%[0-9]+]]:_(s32) = G_CTLZ_ZERO_UNDEF [[UV1]](s64)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[CTLZ_ZERO_UNDEF]](s32), [[CTLZ_ZERO_UNDEF1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    %1:_(<2 x s32>) = G_CTLZ_ZERO_UNDEF %0
+    $vgpr0_vgpr1 = COPY %1
+...
+
+---
+name: ctlz_zero_undef_v2s16_v2s16
+
+body: |
+  bb.0:
+    liveins: $vgpr0
+    ; CHECK-LABEL: name: ctlz_zero_undef_v2s16_v2s16
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; CHECK: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV]](s16)
+    ; CHECK: [[CTLZ_ZERO_UNDEF:%[0-9]+]]:_(s32) = G_CTLZ_ZERO_UNDEF [[ZEXT]](s32)
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CHECK: [[SUB:%[0-9]+]]:_(s32) = G_SUB [[CTLZ_ZERO_UNDEF]], [[C]]
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[SUB]](s32)
+    ; CHECK: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; CHECK: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV1]](s16)
+    ; CHECK: [[CTLZ_ZERO_UNDEF1:%[0-9]+]]:_(s32) = G_CTLZ_ZERO_UNDEF [[ZEXT1]](s32)
+    ; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CHECK: [[SUB1:%[0-9]+]]:_(s32) = G_SUB [[CTLZ_ZERO_UNDEF1]], [[C1]]
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[SUB1]](s32)
+    ; CHECK: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16)
+    ; CHECK: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    %0:_(<2 x s16>) = COPY $vgpr0
+    %1:_(<2 x s16>) = G_CTLZ_ZERO_UNDEF %0
+    $vgpr0 = COPY %1
+...
diff --git a/test/CodeGen/AMDGPU/GlobalISel/legalize-ctlz.mir b/test/CodeGen/AMDGPU/GlobalISel/legalize-ctlz.mir
index acdfc3abd575..5cbdea9d83df 100644
--- a/test/CodeGen/AMDGPU/GlobalISel/legalize-ctlz.mir
+++ b/test/CodeGen/AMDGPU/GlobalISel/legalize-ctlz.mir
@@ -91,3 +91,67 @@ body: |
     %3:_(s32) = G_ZEXT %2
     $vgpr0 = COPY %3
 ...
+
+---
+name: ctlz_v2s32_v2s32
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+    ; CHECK-LABEL: name: ctlz_v2s32_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: [[CTLZ:%[0-9]+]]:_(s32) = G_CTLZ [[UV]](s32)
+    ; CHECK: [[CTLZ1:%[0-9]+]]:_(s32) = G_CTLZ [[UV1]](s32)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[CTLZ]](s32), [[CTLZ1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = G_CTLZ %0
+    $vgpr0_vgpr1 = COPY %1
+...
+
+---
+name: ctlz_v2s32_v2s64
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2_vgpr3
+    ; CHECK-LABEL: name: ctlz_v2s32_v2s64
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; CHECK: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; CHECK: [[CTLZ:%[0-9]+]]:_(s32) = G_CTLZ [[UV]](s64)
+    ; CHECK: [[CTLZ1:%[0-9]+]]:_(s32) = G_CTLZ [[UV1]](s64)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[CTLZ]](s32), [[CTLZ1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    %1:_(<2 x s32>) = G_CTLZ %0
+    $vgpr0_vgpr1 = COPY %1
+...
+
+---
+name: ctlz_v2s16_v2s16
+
+body: |
+  bb.0:
+    liveins: $vgpr0
+    ; CHECK-LABEL: name: ctlz_v2s16_v2s16
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; CHECK: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV]](s16)
+    ; CHECK: [[CTLZ:%[0-9]+]]:_(s32) = G_CTLZ [[ZEXT]](s32)
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CHECK: [[SUB:%[0-9]+]]:_(s32) = G_SUB [[CTLZ]], [[C]]
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[SUB]](s32)
+    ; CHECK: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; CHECK: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV1]](s16)
+    ; CHECK: [[CTLZ1:%[0-9]+]]:_(s32) = G_CTLZ [[ZEXT1]](s32)
+    ; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+    ; CHECK: [[SUB1:%[0-9]+]]:_(s32) = G_SUB [[CTLZ1]], [[C1]]
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[SUB1]](s32)
+    ; CHECK: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16)
+    ; CHECK: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    %0:_(<2 x s16>) = COPY $vgpr0
+    %1:_(<2 x s16>) = G_CTLZ %0
+    $vgpr0 = COPY %1
+...
diff --git a/test/CodeGen/AMDGPU/GlobalISel/legalize-ctpop.mir b/test/CodeGen/AMDGPU/GlobalISel/legalize-ctpop.mir
index d0a272ea2339..ebb59c1bd0d9 100644
--- a/test/CodeGen/AMDGPU/GlobalISel/legalize-ctpop.mir
+++ b/test/CodeGen/AMDGPU/GlobalISel/legalize-ctpop.mir
@@ -89,3 +89,63 @@ body: |
     %3:_(s32) = G_ZEXT %2
     $vgpr0 = COPY %3
 ...
+
+---
+name: ctpop_v2s32_v2s32
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+    ; CHECK-LABEL: name: ctpop_v2s32_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: [[CTPOP:%[0-9]+]]:_(s32) = G_CTPOP [[UV]](s32)
+    ; CHECK: [[CTPOP1:%[0-9]+]]:_(s32) = G_CTPOP [[UV1]](s32)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[CTPOP]](s32), [[CTPOP1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = G_CTPOP %0
+    $vgpr0_vgpr1 = COPY %1
+...
+
+---
+name: ctpop_v2s32_v2s64
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2_vgpr3
+    ; CHECK-LABEL: name: ctpop_v2s32_v2s64
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; CHECK: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; CHECK: [[CTPOP:%[0-9]+]]:_(s32) = G_CTPOP [[UV]](s64)
+    ; CHECK: [[CTPOP1:%[0-9]+]]:_(s32) = G_CTPOP [[UV1]](s64)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[CTPOP]](s32), [[CTPOP1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    %1:_(<2 x s32>) = G_CTPOP %0
+    $vgpr0_vgpr1 = COPY %1
+...
+
+---
+name: ctpop_v2s16_v2s16
+
+body: |
+  bb.0:
+    liveins: $vgpr0
+    ; CHECK-LABEL: name: ctpop_v2s16_v2s16
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; CHECK: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV]](s16)
+    ; CHECK: [[CTPOP:%[0-9]+]]:_(s32) = G_CTPOP [[ZEXT]](s32)
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[CTPOP]](s32)
+    ; CHECK: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; CHECK: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV1]](s16)
+    ; CHECK: [[CTPOP1:%[0-9]+]]:_(s32) = G_CTPOP [[ZEXT1]](s32)
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[CTPOP1]](s32)
+    ; CHECK: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16)
+    ; CHECK: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    %0:_(<2 x s16>) = COPY $vgpr0
+    %1:_(<2 x s16>) = G_CTPOP %0
+    $vgpr0 = COPY %1
+...
diff --git a/test/CodeGen/AMDGPU/GlobalISel/legalize-cttz-zero-undef.mir b/test/CodeGen/AMDGPU/GlobalISel/legalize-cttz-zero-undef.mir
index cb474c59ee10..ba3516be3841 100644
--- a/test/CodeGen/AMDGPU/GlobalISel/legalize-cttz-zero-undef.mir
+++ b/test/CodeGen/AMDGPU/GlobalISel/legalize-cttz-zero-undef.mir
@@ -89,3 +89,63 @@ body: |
     %3:_(s32) = G_ZEXT %2
     $vgpr0 = COPY %3
 ...
+
+---
+name: cttz_zero_undef_v2s32_v2s32
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+    ; CHECK-LABEL: name: cttz_zero_undef_v2s32_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: [[CTTZ_ZERO_UNDEF:%[0-9]+]]:_(s32) = G_CTTZ_ZERO_UNDEF [[UV]](s32)
+    ; CHECK: [[CTTZ_ZERO_UNDEF1:%[0-9]+]]:_(s32) = G_CTTZ_ZERO_UNDEF [[UV1]](s32)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[CTTZ_ZERO_UNDEF]](s32), [[CTTZ_ZERO_UNDEF1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = G_CTTZ_ZERO_UNDEF %0
+    $vgpr0_vgpr1 = COPY %1
+...
+
+---
+name: cttz_zero_undef_v2s32_v2s64
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2_vgpr3
+    ; CHECK-LABEL: name: cttz_zero_undef_v2s32_v2s64
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; CHECK: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; CHECK: [[CTTZ_ZERO_UNDEF:%[0-9]+]]:_(s32) = G_CTTZ_ZERO_UNDEF [[UV]](s64)
+    ; CHECK: [[CTTZ_ZERO_UNDEF1:%[0-9]+]]:_(s32) = G_CTTZ_ZERO_UNDEF [[UV1]](s64)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[CTTZ_ZERO_UNDEF]](s32), [[CTTZ_ZERO_UNDEF1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    %1:_(<2 x s32>) = G_CTTZ_ZERO_UNDEF %0
+    $vgpr0_vgpr1 = COPY %1
+...
+
+---
+name: cttz_zero_undef_v2s16_v2s16
+
+body: |
+  bb.0:
+    liveins: $vgpr0
+    ; CHECK-LABEL: name: cttz_zero_undef_v2s16_v2s16
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; CHECK: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV]](s16)
+    ; CHECK: [[CTTZ_ZERO_UNDEF:%[0-9]+]]:_(s32) = G_CTTZ_ZERO_UNDEF [[ZEXT]](s32)
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[CTTZ_ZERO_UNDEF]](s32)
+    ; CHECK: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; CHECK: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV1]](s16)
+    ; CHECK: [[CTTZ_ZERO_UNDEF1:%[0-9]+]]:_(s32) = G_CTTZ_ZERO_UNDEF [[ZEXT1]](s32)
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[CTTZ_ZERO_UNDEF1]](s32)
+    ; CHECK: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16)
+    ; CHECK: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    %0:_(<2 x s16>) = COPY $vgpr0
+    %1:_(<2 x s16>) = G_CTTZ_ZERO_UNDEF %0
+    $vgpr0 = COPY %1
+...
diff --git a/test/CodeGen/AMDGPU/GlobalISel/legalize-cttz.mir b/test/CodeGen/AMDGPU/GlobalISel/legalize-cttz.mir
index 66e5951198f8..a73f5afcd3da 100644
--- a/test/CodeGen/AMDGPU/GlobalISel/legalize-cttz.mir
+++ b/test/CodeGen/AMDGPU/GlobalISel/legalize-cttz.mir
@@ -91,3 +91,67 @@ body: |
     %3:_(s32) = G_ZEXT %2
     $vgpr0 = COPY %3
 ...
+
+---
+name: cttz_v2s32_v2s32
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+    ; CHECK-LABEL: name: cttz_v2s32_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: [[CTTZ:%[0-9]+]]:_(s32) = G_CTTZ [[UV]](s32)
+    ; CHECK: [[CTTZ1:%[0-9]+]]:_(s32) = G_CTTZ [[UV1]](s32)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[CTTZ]](s32), [[CTTZ1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = G_CTTZ %0
+    $vgpr0_vgpr1 = COPY %1
+...
+
+---
+name: cttz_v2s32_v2s64
+
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2_vgpr3
+    ; CHECK-LABEL: name: cttz_v2s32_v2s64
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    ; CHECK: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
+    ; CHECK: [[CTTZ:%[0-9]+]]:_(s32) = G_CTTZ [[UV]](s64)
+    ; CHECK: [[CTTZ1:%[0-9]+]]:_(s32) = G_CTTZ [[UV1]](s64)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[CTTZ]](s32), [[CTTZ1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    %1:_(<2 x s32>) = G_CTTZ %0
+    $vgpr0_vgpr1 = COPY %1
+...
+
+---
+name: cttz_v2s16_v2s16
+
+body: |
+  bb.0:
+    liveins: $vgpr0
+    ; CHECK-LABEL: name: cttz_v2s16_v2s16
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
+    ; CHECK: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<2 x s16>)
+    ; CHECK: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[UV]](s16)
+    ; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 65536
+    ; CHECK: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[C]]
+    ; CHECK: [[CTTZ:%[0-9]+]]:_(s32) = G_CTTZ [[OR]](s32)
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY [[CTTZ]](s32)
+    ; CHECK: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
+    ; CHECK: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[UV1]](s16)
+    ; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65536
+    ; CHECK: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ZEXT1]], [[C1]]
+    ; CHECK: [[CTTZ1:%[0-9]+]]:_(s32) = G_CTTZ [[OR1]](s32)
+    ; CHECK: [[COPY2:%[0-9]+]]:_(s32) = COPY [[CTTZ1]](s32)
+    ; CHECK: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s32)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16)
+    ; CHECK: $vgpr0 = COPY [[BUILD_VECTOR]](<2 x s16>)
+    %0:_(<2 x s16>) = COPY $vgpr0
+    %1:_(<2 x s16>) = G_CTTZ %0
+    $vgpr0 = COPY %1
+...