AMDGPU/GlobalISel: Scalarize add/sub

llvm-svn: 352167

Author: arsenm

llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp

@@ -1396,6 +1396,7 @@ LegalizerHelper::fewerElementsVector(MachineInstr &MI, unsigned TypeIdx,
     return Legalized;
   }
   case TargetOpcode::G_ADD:
+  case TargetOpcode::G_SUB:
   case TargetOpcode::G_MUL:
   case TargetOpcode::G_SMULH:
   case TargetOpcode::G_UMULH:

llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp

@@ -93,12 +93,10 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST,
 
   setAction({G_BRCOND, S1}, Legal);
 
-  setAction({G_ADD, S32}, Legal);
   setAction({G_ASHR, S32}, Legal);
   setAction({G_ASHR, 1, S32}, Legal);
-  setAction({G_SUB, S32}, Legal);
 
-  getActionDefinitionsBuilder({G_MUL, G_UMULH, G_SMULH})
+  getActionDefinitionsBuilder({G_ADD, G_SUB, G_MUL, G_UMULH, G_SMULH})
     .legalFor({S32})
     .scalarize(0);
 

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-add.mir

@@ -2,17 +2,39 @@
 # RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -O0 -run-pass=legalizer %s -o - | FileCheck %s
 
 ---
-name:            test_add
+name: test_add_s32
 body: |
   bb.0:
     liveins: $vgpr0, $vgpr1
 
-    ; CHECK-LABEL: name: test_add
+    ; CHECK-LABEL: name: test_add_s32
     ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
     ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
     ; CHECK: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[COPY]], [[COPY1]]
+    ; CHECK: $vgpr0 = COPY [[ADD]](s32)
     %0:_(s32) = COPY $vgpr0
     %1:_(s32) = COPY $vgpr1
     %2:_(s32) = G_ADD %0, %1
     $vgpr0 = COPY %2
 ...
+
+---
+name: test_add_v2s32
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+    ; CHECK-LABEL: name: test_add_v2s32
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; CHECK: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; CHECK: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[UV]], [[UV2]]
+    ; CHECK: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[UV1]], [[UV3]]
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[ADD]](s32), [[ADD1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    %2:_(<2 x s32>) = G_ADD %0, %1
+    $vgpr0_vgpr1 = COPY %2
+...

llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-sub.mir

@@ -0,0 +1,40 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -O0 -run-pass=legalizer %s -o - | FileCheck %s
+
+---
+name: test_sub_s32
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+
+    ; CHECK-LABEL: name: test_sub_s32
+    ; CHECK: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
+    ; CHECK: [[SUB:%[0-9]+]]:_(s32) = G_SUB [[COPY]], [[COPY1]]
+    ; CHECK: $vgpr0 = COPY [[SUB]](s32)
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(s32) = G_SUB %0, %1
+    $vgpr0 = COPY %2
+...
+
+---
+name: test_sub_v2s32
+body: |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
+
+    ; CHECK-LABEL: name: test_sub_v2s32
+    ; CHECK: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    ; CHECK: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
+    ; CHECK: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
+    ; CHECK: [[SUB:%[0-9]+]]:_(s32) = G_SUB [[UV]], [[UV2]]
+    ; CHECK: [[SUB1:%[0-9]+]]:_(s32) = G_SUB [[UV1]], [[UV3]]
+    ; CHECK: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[SUB]](s32), [[SUB1]](s32)
+    ; CHECK: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
+    %0:_(<2 x s32>) = COPY $vgpr0_vgpr1
+    %1:_(<2 x s32>) = COPY $vgpr2_vgpr3
+    %2:_(<2 x s32>) = G_SUB %0, %1
+    $vgpr0_vgpr1 = COPY %2
+...