AMDGPU/GlobalISel: Private loads always use VGPRs

llvm-svn: 373414

diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
index 49a4c7b..67a7f0a 100644 (file)
--- a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
@@ -447,8 +447,9 @@ AMDGPURegisterBankInfo::getInstrAlternativeMappings(
     unsigned PtrSize = PtrTy.getSizeInBits();
     unsigned AS = PtrTy.getAddressSpace();
     LLT LoadTy = MRI.getType(MI.getOperand(0).getReg());
-    if (isInstrUniformNonExtLoadAlign4(MI) &&
-        (AS != AMDGPUAS::LOCAL_ADDRESS && AS != AMDGPUAS::REGION_ADDRESS)) {
+    if ((AS != AMDGPUAS::LOCAL_ADDRESS && AS != AMDGPUAS::REGION_ADDRESS &&
+         AS != AMDGPUAS::PRIVATE_ADDRESS) &&
+        isInstrUniformNonExtLoadAlign4(MI)) {
       const InstructionMapping &SSMapping = getInstructionMapping(
           1, 1, getOperandsMapping(
                     {AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size),
@@ -1853,8 +1854,9 @@ AMDGPURegisterBankInfo::getInstrMappingForLoad(const MachineInstr &MI) const {
   const ValueMapping *ValMapping;
   const ValueMapping *PtrMapping;
 
-  if (isInstrUniformNonExtLoadAlign4(MI) &&
-      (AS != AMDGPUAS::LOCAL_ADDRESS && AS != AMDGPUAS::REGION_ADDRESS)) {
+  if ((AS != AMDGPUAS::LOCAL_ADDRESS && AS != AMDGPUAS::REGION_ADDRESS &&
+       AS != AMDGPUAS::PRIVATE_ADDRESS) &&
+      isInstrUniformNonExtLoadAlign4(MI)) {
     // We have a uniform instruction so we want to use an SMRD load
     ValMapping = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size);
     PtrMapping = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, PtrSize);

diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-load.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-load.mir
index d129383..49ac13f 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-load.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-load.mir
@@ -68,6 +68,7 @@
   define amdgpu_kernel void @load_constant_i32_uniform_align4() {ret void}
   define amdgpu_kernel void @load_constant_i32_uniform_align2() {ret void}
   define amdgpu_kernel void @load_constant_i32_uniform_align1() {ret void}
+  define amdgpu_kernel void @load_private_uniform_sgpr_i32() {ret void}
 
   declare i32 @llvm.amdgcn.workitem.id.x() #0
   attributes #0 = { nounwind readnone }
@@ -635,3 +636,19 @@ body: |
     %0:_(p4) = COPY $sgpr0_sgpr1
     %1:_(s32) = G_LOAD %0 :: (load 4, addrspace 4, align 1)
 ...
+
+---
+name: load_private_uniform_sgpr_i32
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0
+
+    ; CHECK-LABEL: name: load_private_uniform_sgpr_i32
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(p5) = COPY $sgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(p5) = COPY [[COPY]](p5)
+    ; CHECK: [[LOAD:%[0-9]+]]:vgpr(s32) = G_LOAD [[COPY1]](p5) :: (load 4, addrspace 5)
+    %0:_(p5) = COPY $sgpr0
+    %1:_(s32) = G_LOAD %0 :: (load 4, addrspace 5, align 4)
+...