From e6cebd0d69ec29aa41f560d3d20a3da452146697 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault@amd.com>
Date: Fri, 25 Jan 2019 04:37:33 +0000
Subject: [PATCH] GlobalISel: fewerElementsVector for more cast types

llvm-svn: 352166
---
 .../CodeGen/GlobalISel/LegalizerHelper.cpp    |  5 +++
 .../lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp |  9 ++--
 .../AMDGPU/GlobalISel/legalize-fptosi.mir     | 42 +++++++++++++++++++
 .../AMDGPU/GlobalISel/legalize-fptoui.mir     | 35 ++++++++++++++++
 .../AMDGPU/GlobalISel/legalize-fptrunc.mir    | 39 +++++++++++++++++
 .../AMDGPU/GlobalISel/legalize-sitofp.mir     | 35 ++++++++++++++++
 6 files changed, 162 insertions(+), 3 deletions(-)

diff --git a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
index 5be14018fb9f..d616d7311f82 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
@@ -1511,6 +1511,11 @@ LegalizerHelper::fewerElementsVector(MachineInstr &MI, unsigned TypeIdx,
   case TargetOpcode::G_SEXT:
   case TargetOpcode::G_ANYEXT:
   case TargetOpcode::G_FPEXT:
+  case TargetOpcode::G_FPTRUNC:
+  case TargetOpcode::G_SITOFP:
+  case TargetOpcode::G_UITOFP:
+  case TargetOpcode::G_FPTOSI:
+  case TargetOpcode::G_FPTOUI:
     return fewerElementsVectorCasts(MI, TypeIdx, NarrowTy);
   }
 }
diff --git a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
index 5ad9a43bb960..e3f34404e3d9 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -148,7 +148,8 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST,
       .clampScalar(0, S32, S64);
 
   getActionDefinitionsBuilder(G_FPTRUNC)
-    .legalFor({{S32, S64}, {S16, S32}});
+    .legalFor({{S32, S64}, {S16, S32}})
+    .scalarize(0);
 
   getActionDefinitionsBuilder(G_FPEXT)
     .legalFor({{S64, S32}, {S32, S16}})
@@ -171,10 +172,12 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST,
     .scalarize(0);
 
   getActionDefinitionsBuilder({G_SITOFP, G_UITOFP})
-    .legalFor({{S32, S32}, {S64, S32}});
+    .legalFor({{S32, S32}, {S64, S32}})
+    .scalarize(0);
 
   getActionDefinitionsBuilder({G_FPTOSI, G_FPTOUI})
-    .legalFor({{S32, S32}, {S32, S64}});
+    .legalFor({{S32, S32}, {S32, S64}})
+    .scalarize(0);
 
   getActionDefinitionsBuilder({G_INTRINSIC_TRUNC, G_INTRINSIC_ROUND})
     .legalFor({S32, S64});
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptosi.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptosi.mir
index 60c2e2aecaaf..2da5ffd1d034 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptosi.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptosi.mir
@@ -9,8 +9,11 @@ body: |
 
     ; CHECK-LABEL: name: test_fptosi_s32_to_s32
     ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; CHECK: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[COPY]](s32)
+    ; CHECK: $vgpr0 = COPY [[FPTOSI]](s32)
     %0:_(s32) = COPY $vgpr0
     %1:_(s32) = G_FPTOSI %0
+    $vgpr0 = COPY %1
 ...
 
 ---
@@ -21,6 +24,45 @@ body: |
 
     ; CHECK-LABEL: name: test_fptosi_s64_to_s32
     ; CHECK: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
+    ; CHECK: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[COPY]](s64)
+    ; CHECK: $vgpr0 = COPY [[FPTOSI]](s32)
     %0:_(s64) = COPY $vgpr0_vgpr1
     %1:_(s32) = G_FPTOSI %0
+    $vgpr0 = COPY %1
+...
+
+---
+name: test_fptosi_v2s32_to_v2s32
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+
+    ; CHECK-LABEL: name: test_fptosi_v2s32_to_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: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[UV]](s32)
+    ; CHECK: [[FPTOSI1:%[0-9]+]]:_(s32) = G_FPTOSI [[UV1]](s32)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FPTOSI]](s32), [[FPTOSI1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = G_FPTOSI %0
+    $vgpr0_vgpr1 = COPY %1
+...
+
+---
+name: test_fptosi_v2s64_to_v2s32
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2_vgpr3
+
+    ; CHECK-LABEL: name: test_fptosi_v2s64_to_v2s32
+    ; 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: [[FPTOSI:%[0-9]+]]:_(s32) = G_FPTOSI [[UV]](s64)
+    ; CHECK: [[FPTOSI1:%[0-9]+]]:_(s32) = G_FPTOSI [[UV1]](s64)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FPTOSI]](s32), [[FPTOSI1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    %1:_(<2 x s32>) = G_FPTOSI %0
+    $vgpr0_vgpr1 = COPY %1
 ...
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptoui.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptoui.mir
index c77b8ed21514..f82c72bd2332 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptoui.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptoui.mir
@@ -31,3 +31,38 @@ body: |
     $vgpr0 = COPY %1
 ...
 
+---
+name: test_fptoui_v2s32_to_v2s32
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+
+    ; CHECK-LABEL: name: test_fptoui_v2s32_to_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: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[UV]](s32)
+    ; CHECK: [[FPTOUI1:%[0-9]+]]:_(s32) = G_FPTOUI [[UV1]](s32)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FPTOUI]](s32), [[FPTOUI1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = G_FPTOUI %0
+    $vgpr0_vgpr1 = COPY %1
+...
+
+---
+name: test_fptoui_v2s64_to_v2s32
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2_vgpr3
+
+    ; CHECK-LABEL: name: test_fptoui_v2s64_to_v2s32
+    ; 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: [[FPTOUI:%[0-9]+]]:_(s32) = G_FPTOUI [[UV]](s64)
+    ; CHECK: [[FPTOUI1:%[0-9]+]]:_(s32) = G_FPTOUI [[UV1]](s64)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FPTOUI]](s32), [[FPTOUI1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    %1:_(<2 x s32>) = G_FPTOUI %0
+    $vgpr0_vgpr1 = COPY %1
+...
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptrunc.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptrunc.mir
index f2ed866424e4..5a66e8a6c2e0 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptrunc.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-fptrunc.mir
@@ -32,3 +32,42 @@ body: |
     %2:_(s32) = G_ANYEXT %1
     $vgpr0 = COPY %2
 ...
+
+---
+name: test_fptrunc_v2s64_to_v2s32
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1_vgpr2_vgpr3
+
+    ; CHECK-LABEL: name: test_fptrunc_v2s64_to_v2s32
+    ; 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: [[FPTRUNC:%[0-9]+]]:_(s32) = G_FPTRUNC [[UV]](s64)
+    ; CHECK: [[FPTRUNC1:%[0-9]+]]:_(s32) = G_FPTRUNC [[UV1]](s64)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FPTRUNC]](s32), [[FPTRUNC1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
+    %1:_(<2 x s32>) = G_FPTRUNC %0
+    $vgpr0_vgpr1 = COPY %1
+...
+
+---
+name: test_fptrunc_v2s32_to_v2s16
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+
+    ; CHECK-LABEL: name: test_fptrunc_v2s32_to_v2s16
+    ; 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: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[UV]](s32)
+    ; CHECK: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[UV1]](s32)
+    ; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16)
+    ; CHECK: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC1]](s16)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[ANYEXT]](s32), [[ANYEXT1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s16>) = G_FPTRUNC %0
+    %2:_(<2 x s32>) = G_ANYEXT %1
+    $vgpr0_vgpr1 = COPY %2
+...
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-sitofp.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-sitofp.mir
index aa86c5f02533..6a45e73a18c3 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-sitofp.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-sitofp.mir
@@ -31,3 +31,38 @@ body: |
     $vgpr0_vgpr1 = COPY %1
 ...
 
+---
+name: test_sitofp_v2s32_to_v2s32
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+
+    ; CHECK-LABEL: name: test_sitofp_v2s32_to_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: [[SITOFP:%[0-9]+]]:_(s32) = G_SITOFP [[UV]](s32)
+    ; CHECK: [[SITOFP1:%[0-9]+]]:_(s32) = G_SITOFP [[UV1]](s32)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[SITOFP]](s32), [[SITOFP1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = G_SITOFP %0
+    $vgpr0_vgpr1 = COPY %1
+...
+
+---
+name: test_sitofp_v2s32_to_v2s64
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+
+    ; CHECK-LABEL: name: test_sitofp_v2s32_to_v2s64
+    ; 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: [[SITOFP:%[0-9]+]]:_(s64) = G_SITOFP [[UV]](s32)
+    ; CHECK: [[SITOFP1:%[0-9]+]]:_(s64) = G_SITOFP [[UV1]](s32)
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[SITOFP]](s64), [[SITOFP1]](s64)
+    ; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s64>) = G_SITOFP %0
+    $vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
+...