AMDGPU/GlobalISel: RegBankSelect for G_ZEXTLOAD/G_SEXTLOAD

llvm-svn: 371536

diff --git a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
index ff292f5..aa65fb6 100644 (file)
--- a/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -780,7 +780,9 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST_,
                                                   {S32, LocalPtr, 8, 8},
                                                   {S32, LocalPtr, 16, 16},
                                                   {S32, PrivatePtr, 8, 8},
-                                                  {S32, PrivatePtr, 16, 16}});
+                                                  {S32, PrivatePtr, 16, 16},
+                                                  {S32, ConstantPtr, 8, 8},
+                                                  {S32, ConstantPtr, 16, 2 * 8}});
   if (ST.hasFlatAddressSpace()) {
     ExtLoads.legalForTypesWithMemDesc(
         {{S32, FlatPtr, 8, 8}, {S32, FlatPtr, 16, 16}});

diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
index c5ec496..60f0d37 100644 (file)
--- a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
@@ -421,7 +421,9 @@ AMDGPURegisterBankInfo::getInstrAlternativeMappings(
     AltMappings.push_back(&VSMapping);
     break;
   }
-  case TargetOpcode::G_LOAD: {
+  case TargetOpcode::G_LOAD:
+  case TargetOpcode::G_ZEXTLOAD:
+  case TargetOpcode::G_SEXTLOAD: {
     unsigned Size = getSizeInBits(MI.getOperand(0).getReg(), MRI, *TRI);
     LLT PtrTy = MRI.getType(MI.getOperand(1).getReg());
     unsigned PtrSize = PtrTy.getSizeInBits();
@@ -1427,7 +1429,9 @@ void AMDGPURegisterBankInfo::applyMappingImpl(
     }
     break;
   }
-  case AMDGPU::G_LOAD: {
+  case AMDGPU::G_LOAD:
+  case AMDGPU::G_ZEXTLOAD:
+  case AMDGPU::G_SEXTLOAD: {
     if (applyMappingWideLoad(MI, OpdMapper, MRI))
       return;
     break;
@@ -2321,6 +2325,8 @@ AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
   }
 
   case AMDGPU::G_LOAD:
+  case AMDGPU::G_ZEXTLOAD:
+  case AMDGPU::G_SEXTLOAD:
     return getInstrMappingForLoad(MI);
 
   case AMDGPU::G_ATOMICRMW_XCHG:

diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sextload.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sextload.mir
new file mode 100644 (file)
index 0000000..70857f8
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-sextload.mir
@@ -0,0 +1,98 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -march=amdgcn -run-pass=regbankselect  -regbankselect-fast -verify-machineinstrs -o - %s | FileCheck %s
+# RUN: llc -march=amdgcn -run-pass=regbankselect  -regbankselect-greedy -verify-machineinstrs -o - %s | FileCheck %s
+
+---
+name: sextload_constant_i8_to_i32_uniform
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0_sgpr1
+    ; CHECK-LABEL: name: sextload_constant_i8_to_i32_uniform
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(p4) = COPY $sgpr0_sgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(p4) = COPY [[COPY]](p4)
+    ; CHECK: [[SEXTLOAD:%[0-9]+]]:vgpr(s32) = G_SEXTLOAD [[COPY1]](p4) :: (load 1, addrspace 4)
+    %0:_(p4) = COPY $sgpr0_sgpr1
+    %1:_(s32) = G_SEXTLOAD %0 :: (load 1, addrspace 4, align 1)
+...
+
+---
+name: sextload_global_i8_to_i32_uniform
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0_sgpr1
+
+    ; CHECK-LABEL: name: sextload_global_i8_to_i32_uniform
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(p4) = COPY $sgpr0_sgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(p4) = COPY [[COPY]](p4)
+    ; CHECK: [[SEXTLOAD:%[0-9]+]]:vgpr(s32) = G_SEXTLOAD [[COPY1]](p4) :: (load 1, addrspace 1)
+    %0:_(p4) = COPY $sgpr0_sgpr1
+    %1:_(s32) = G_SEXTLOAD %0 :: (load 1, addrspace 1, align 1)
+...
+
+---
+name: sextload_constant_i16_to_i32_uniform
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0_sgpr1
+
+    ; CHECK-LABEL: name: sextload_constant_i16_to_i32_uniform
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(p4) = COPY $sgpr0_sgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(p4) = COPY [[COPY]](p4)
+    ; CHECK: [[SEXTLOAD:%[0-9]+]]:vgpr(s32) = G_SEXTLOAD [[COPY1]](p4) :: (load 2, addrspace 4)
+    %0:_(p4) = COPY $sgpr0_sgpr1
+    %1:_(s32) = G_SEXTLOAD %0 :: (load 2, addrspace 4, align 2)
+...
+
+---
+name: sextload_global_i16_to_i32_uniform
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0_sgpr1
+
+    ; CHECK-LABEL: name: sextload_global_i16_to_i32_uniform
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(p4) = COPY $sgpr0_sgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(p4) = COPY [[COPY]](p4)
+    ; CHECK: [[SEXTLOAD:%[0-9]+]]:vgpr(s32) = G_SEXTLOAD [[COPY1]](p4) :: (load 2, addrspace 1)
+    %0:_(p4) = COPY $sgpr0_sgpr1
+    %1:_(s32) = G_SEXTLOAD %0 :: (load 2, addrspace 1, align 2)
+...
+
+---
+name: sextload_local_i8_to_i32_uniform
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0
+    ; CHECK-LABEL: name: sextload_local_i8_to_i32_uniform
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(p3) = COPY $sgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(p3) = COPY [[COPY]](p3)
+    ; CHECK: [[SEXTLOAD:%[0-9]+]]:vgpr(s32) = G_SEXTLOAD [[COPY1]](p3) :: (load 1, addrspace 3)
+    %0:_(p3) = COPY $sgpr0
+    %1:_(s32) = G_SEXTLOAD %0 :: (load 1, addrspace 3, align 1)
+...
+
+---
+name: sextload_local_i16_to_i32_uniform
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0
+
+    ; CHECK-LABEL: name: sextload_local_i16_to_i32_uniform
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(p3) = COPY $sgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(p3) = COPY [[COPY]](p3)
+    ; CHECK: [[SEXTLOAD:%[0-9]+]]:vgpr(s32) = G_SEXTLOAD [[COPY1]](p3) :: (load 2, addrspace 3)
+    %0:_(p3) = COPY $sgpr0
+    %1:_(s32) = G_SEXTLOAD %0 :: (load 2, addrspace 3, align 2)
+...
+!

diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-zextload.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-zextload.mir
new file mode 100644 (file)
index 0000000..3ace0d3
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-zextload.mir
@@ -0,0 +1,97 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -march=amdgcn -run-pass=regbankselect  -regbankselect-fast -verify-machineinstrs -o - %s | FileCheck %s
+# RUN: llc -march=amdgcn -run-pass=regbankselect  -regbankselect-greedy -verify-machineinstrs -o - %s | FileCheck %s
+
+---
+name: zextload_constant_i8_to_i32_uniform
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0_sgpr1
+    ; CHECK-LABEL: name: zextload_constant_i8_to_i32_uniform
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(p4) = COPY $sgpr0_sgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(p4) = COPY [[COPY]](p4)
+    ; CHECK: [[ZEXTLOAD:%[0-9]+]]:vgpr(s32) = G_ZEXTLOAD [[COPY1]](p4) :: (load 1, addrspace 4)
+    %0:_(p4) = COPY $sgpr0_sgpr1
+    %1:_(s32) = G_ZEXTLOAD %0 :: (load 1, addrspace 4, align 1)
+...
+
+---
+name: zextload_global_i8_to_i32_uniform
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0_sgpr1
+
+    ; CHECK-LABEL: name: zextload_global_i8_to_i32_uniform
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(p4) = COPY $sgpr0_sgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(p4) = COPY [[COPY]](p4)
+    ; CHECK: [[ZEXTLOAD:%[0-9]+]]:vgpr(s32) = G_ZEXTLOAD [[COPY1]](p4) :: (load 1, addrspace 1)
+    %0:_(p4) = COPY $sgpr0_sgpr1
+    %1:_(s32) = G_ZEXTLOAD %0 :: (load 1, addrspace 1, align 1)
+...
+
+---
+name: zextload_constant_i16_to_i32_uniform
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0_sgpr1
+
+    ; CHECK-LABEL: name: zextload_constant_i16_to_i32_uniform
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(p4) = COPY $sgpr0_sgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(p4) = COPY [[COPY]](p4)
+    ; CHECK: [[ZEXTLOAD:%[0-9]+]]:vgpr(s32) = G_ZEXTLOAD [[COPY1]](p4) :: (load 2, addrspace 4)
+    %0:_(p4) = COPY $sgpr0_sgpr1
+    %1:_(s32) = G_ZEXTLOAD %0 :: (load 2, addrspace 4, align 2)
+...
+
+---
+name: zextload_global_i16_to_i32_uniform
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0_sgpr1
+
+    ; CHECK-LABEL: name: zextload_global_i16_to_i32_uniform
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(p4) = COPY $sgpr0_sgpr1
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(p4) = COPY [[COPY]](p4)
+    ; CHECK: [[ZEXTLOAD:%[0-9]+]]:vgpr(s32) = G_ZEXTLOAD [[COPY1]](p4) :: (load 2, addrspace 1)
+    %0:_(p4) = COPY $sgpr0_sgpr1
+    %1:_(s32) = G_ZEXTLOAD %0 :: (load 2, addrspace 1, align 2)
+...
+
+---
+name: zextload_local_i8_to_i32_uniform
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0
+    ; CHECK-LABEL: name: zextload_local_i8_to_i32_uniform
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(p3) = COPY $sgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(p3) = COPY [[COPY]](p3)
+    ; CHECK: [[ZEXTLOAD:%[0-9]+]]:vgpr(s32) = G_ZEXTLOAD [[COPY1]](p3) :: (load 1, addrspace 3)
+    %0:_(p3) = COPY $sgpr0
+    %1:_(s32) = G_ZEXTLOAD %0 :: (load 1, addrspace 3, align 1)
+...
+
+---
+name: zextload_local_i16_to_i32_uniform
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0
+
+    ; CHECK-LABEL: name: zextload_local_i16_to_i32_uniform
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(p3) = COPY $sgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(p3) = COPY [[COPY]](p3)
+    ; CHECK: [[ZEXTLOAD:%[0-9]+]]:vgpr(s32) = G_ZEXTLOAD [[COPY1]](p3) :: (load 2, addrspace 3)
+    %0:_(p3) = COPY $sgpr0
+    %1:_(s32) = G_ZEXTLOAD %0 :: (load 2, addrspace 3, align 2)
+...