AMDGPU/GlobalISel: Select easy cases for G_BUILD_VECTOR

llvm-svn: 366087

2 files changed, 156 insertions, 0 deletions

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
index 5b2dfc8ff50..6b827db6faa 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUInstructionSelector.cpp
@@ -341,6 +341,9 @@ bool AMDGPUInstructionSelector::selectG_MERGE_VALUES(MachineInstr &MI) const {
   LLT SrcTy = MRI.getType(MI.getOperand(1).getReg());
 
   const unsigned SrcSize = SrcTy.getSizeInBits();
+  if (SrcSize < 32)
+    return false;
+
   const DebugLoc &DL = MI.getDebugLoc();
   const RegisterBank *DstBank = RBI.getRegBank(DstReg, MRI, TRI);
   const unsigned DstSize = DstTy.getSizeInBits();
@@ -1235,6 +1238,7 @@ bool AMDGPUInstructionSelector::select(MachineInstr &I,
   case TargetOpcode::G_EXTRACT:
     return selectG_EXTRACT(I);
   case TargetOpcode::G_MERGE_VALUES:
+  case TargetOpcode::G_BUILD_VECTOR:
   case TargetOpcode::G_CONCAT_VECTORS:
     return selectG_MERGE_VALUES(I);
   case TargetOpcode::G_UNMERGE_VALUES:
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-build-vector.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-build-vector.mir
new file mode 100644
index 00000000000..d97118bf148
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/inst-select-build-vector.mir
@@ -0,0 +1,152 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -run-pass=instruction-select -verify-machineinstrs  -o - %s | FileCheck -check-prefix=GCN  %s
+
+---
+name: test_build_vector_v_v2s32_v_s32_v_s32
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+
+    ; GCN-LABEL: name: test_build_vector_v_v2s32_v_s32_v_s32
+    ; GCN: liveins: $vgpr0, $vgpr1
+    ; GCN: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; GCN: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1
+    ; GCN: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
+    ; GCN: S_ENDPGM 0, implicit [[REG_SEQUENCE]]
+    %0:vgpr(s32) = COPY $vgpr0
+    %1:vgpr(s32) = COPY $vgpr1
+    %2:vgpr(<2 x s32>) = G_BUILD_VECTOR %0, %1
+    S_ENDPGM 0, implicit %2
+...
+
+---
+name: test_build_vector_v_v2s32_s_s32_v_s32
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0, $vgpr0
+
+    ; GCN-LABEL: name: test_build_vector_v_v2s32_s_s32_v_s32
+    ; GCN: liveins: $sgpr0, $vgpr0
+    ; GCN: [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr0
+    ; GCN: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; GCN: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
+    ; GCN: S_ENDPGM 0, implicit [[REG_SEQUENCE]]
+    %0:sgpr(s32) = COPY $sgpr0
+    %1:vgpr(s32) = COPY $vgpr0
+    %2:vgpr(<2 x s32>) = G_BUILD_VECTOR %0, %1
+    S_ENDPGM 0, implicit %2
+...
+
+---
+name: test_build_vector_v_v2s32_v_s32_s_s32
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0, $vgpr0
+
+    ; GCN-LABEL: name: test_build_vector_v_v2s32_v_s32_s_s32
+    ; GCN: liveins: $sgpr0, $vgpr0
+    ; GCN: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0
+    ; GCN: [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr0
+    ; GCN: [[REG_SEQUENCE:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
+    ; GCN: S_ENDPGM 0, implicit [[REG_SEQUENCE]]
+    %0:vgpr(s32) = COPY $vgpr0
+    %1:sgpr(s32) = COPY $sgpr0
+    %2:vgpr(<2 x s32>) = G_BUILD_VECTOR %0, %1
+    S_ENDPGM 0, implicit %2
+...
+
+---
+name: test_build_vector_s_v2s32_s_s32_s_s32
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0, $sgpr1
+
+    ; GCN-LABEL: name: test_build_vector_s_v2s32_s_s32_s_s32
+    ; GCN: liveins: $sgpr0, $sgpr1
+    ; GCN: [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr0
+    ; GCN: [[COPY1:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr1
+    ; GCN: [[REG_SEQUENCE:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1
+    ; GCN: S_ENDPGM 0, implicit [[REG_SEQUENCE]]
+    %0:sgpr(s32) = COPY $sgpr0
+    %1:sgpr(s32) = COPY $sgpr1
+    %2:sgpr(<2 x s32>) = G_BUILD_VECTOR %0, %1
+    S_ENDPGM 0, implicit %2
+...
+
+---
+name: test_build_vector_s_v2s32_undef_s_s32_s_s32
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0
+
+    ; GCN-LABEL: name: test_build_vector_s_v2s32_undef_s_s32_s_s32
+    ; GCN: liveins: $sgpr0
+    ; GCN: [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr0
+    ; GCN: [[REG_SEQUENCE:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE undef %2:sreg_32_xm0, %subreg.sub0, [[COPY]], %subreg.sub1
+    ; GCN: S_ENDPGM 0, implicit [[REG_SEQUENCE]]
+    %1:sgpr(s32) = COPY $sgpr0
+    %2:sgpr(<2 x s32>) = G_BUILD_VECTOR undef %0:sgpr(s32), %1
+    S_ENDPGM 0, implicit %2
+...
+
+---
+name: test_build_vector_s_v2s32_s_s32_undef_s_s32
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0
+
+    ; GCN-LABEL: name: test_build_vector_s_v2s32_s_s32_undef_s_s32
+    ; GCN: liveins: $sgpr0
+    ; GCN: [[COPY:%[0-9]+]]:sreg_32_xm0 = COPY $sgpr0
+    ; GCN: [[REG_SEQUENCE:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[COPY]], %subreg.sub0, undef %2:sreg_32_xm0, %subreg.sub1
+    ; GCN: S_ENDPGM 0, implicit [[REG_SEQUENCE]]
+    %0:sgpr(s32) = COPY $sgpr0
+    %2:sgpr(<2 x s32>) = G_BUILD_VECTOR %0, undef %1:sgpr(s32),
+    S_ENDPGM 0, implicit %2
+...
+
+---
+name: test_build_vector_s_v2s64_s_s64_s_s64
+legalized:       true
+regBankSelected: true
+tracksRegLiveness: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0_sgpr1, $sgpr2_sgpr3
+
+    ; GCN-LABEL: name: test_build_vector_s_v2s64_s_s64_s_s64
+    ; GCN: liveins: $sgpr0_sgpr1, $sgpr2_sgpr3
+    ; GCN: [[COPY:%[0-9]+]]:sreg_64_xexec = COPY $sgpr0_sgpr1
+    ; GCN: [[COPY1:%[0-9]+]]:sreg_64_xexec = COPY $sgpr2_sgpr3
+    ; GCN: [[REG_SEQUENCE:%[0-9]+]]:sreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0_sub1, [[COPY1]], %subreg.sub2_sub3
+    ; GCN: $sgpr0_sgpr1_sgpr2_sgpr3 = COPY [[REG_SEQUENCE]]
+    %0:sgpr(s64) = COPY $sgpr0_sgpr1
+    %1:sgpr(s64) = COPY $sgpr2_sgpr3
+    %4:sgpr(<2 x s64>) = G_BUILD_VECTOR %0, %1
+    $sgpr0_sgpr1_sgpr2_sgpr3 = COPY %4
+...