From: Matt Arsenault <Matthew.Arsenault@amd.com>
Date: Tue, 8 Jan 2019 01:03:58 +0000 (+0000)
Subject: AMDGPU/GlobalISel: RegBankSelect for add/sub with carry out
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=9054ffcfc837f6a649db81746e1b76379355fb28;p=llvm

AMDGPU/GlobalISel: RegBankSelect for add/sub with carry out

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@350589 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp b/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
index 9a27401fa4a..afdc2588f0a 100644
--- a/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
+++ b/lib/Target/AMDGPU/AMDGPULegalizerInfo.cpp
@@ -94,6 +94,9 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST,
   getActionDefinitionsBuilder({G_AND, G_OR, G_XOR})
     .legalFor({S32, S1, S64, V2S32});
 
+  getActionDefinitionsBuilder({G_UADDO, G_SADDO, G_USUBO, G_SSUBO})
+    .legalFor({{S32, S1}});
+
   setAction({G_BITCAST, V2S16}, Legal);
   setAction({G_BITCAST, 1, S32}, Legal);
 
diff --git a/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp b/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
index fd3c9489d0f..01ccfb39626 100644
--- a/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
+++ b/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
@@ -216,9 +216,13 @@ bool AMDGPURegisterBankInfo::isSALUMapping(const MachineInstr &MI) const {
     if (!MI.getOperand(i).isReg())
       continue;
     unsigned Reg = MI.getOperand(i).getReg();
-    const RegisterBank *Bank = getRegBank(Reg, MRI, *TRI);
-    if (Bank && Bank->getID() != AMDGPU::SGPRRegBankID)
-      return false;
+    if (const RegisterBank *Bank = getRegBank(Reg, MRI, *TRI)) {
+      if (Bank->getID() == AMDGPU::VGPRRegBankID)
+        return false;
+
+      assert(Bank->getID() == AMDGPU::SGPRRegBankID ||
+             Bank->getID() == AMDGPU::SCCRegBankID);
+    }
   }
   return true;
 }
@@ -231,7 +235,8 @@ AMDGPURegisterBankInfo::getDefaultMappingSOP(const MachineInstr &MI) const {
 
   for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) {
     unsigned Size = getSizeInBits(MI.getOperand(i).getReg(), MRI, *TRI);
-    OpdsMapping[i] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size);
+    unsigned BankID = Size == 1 ? AMDGPU::SCCRegBankID : AMDGPU::SGPRRegBankID;
+    OpdsMapping[i] = AMDGPU::getValueMapping(BankID, Size);
   }
   return getInstructionMapping(1, 1, getOperandsMapping(OpdsMapping),
                                MI.getNumOperands());
@@ -252,7 +257,11 @@ AMDGPURegisterBankInfo::getDefaultMappingVOP(const MachineInstr &MI) const {
 
   unsigned Reg1 = MI.getOperand(OpdIdx).getReg();
   unsigned Size1 = getSizeInBits(Reg1, MRI, *TRI);
-  unsigned Bank1 = getRegBankID(Reg1, MRI, *TRI);
+
+  unsigned DefaultBankID = Size1 == 1 ?
+    AMDGPU::SGPRRegBankID : AMDGPU::VGPRRegBankID;
+  unsigned Bank1 = getRegBankID(Reg1, MRI, *TRI, DefaultBankID);
+
   OpdsMapping[OpdIdx++] = AMDGPU::getValueMapping(Bank1, Size1);
 
   for (unsigned e = MI.getNumOperands(); OpdIdx != e; ++OpdIdx) {
@@ -359,6 +368,10 @@ AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
   case AMDGPU::G_SUB:
   case AMDGPU::G_MUL:
   case AMDGPU::G_SHL:
+  case AMDGPU::G_UADDO:
+  case AMDGPU::G_SADDO:
+  case AMDGPU::G_USUBO:
+  case AMDGPU::G_SSUBO:
     if (isSALUMapping(MI))
       return getDefaultMappingSOP(MI);
     LLVM_FALLTHROUGH;
diff --git a/test/CodeGen/AMDGPU/GlobalISel/regbankselect-saddo.mir b/test/CodeGen/AMDGPU/GlobalISel/regbankselect-saddo.mir
new file mode 100644
index 00000000000..41399adfa15
--- /dev/null
+++ b/test/CodeGen/AMDGPU/GlobalISel/regbankselect-saddo.mir
@@ -0,0 +1,68 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -march=amdgcn -mcpu=fiji -run-pass=regbankselect %s -verify-machineinstrs -o - -regbankselect-fast | FileCheck %s
+# RUN: llc -march=amdgcn -mcpu=fiji -run-pass=regbankselect %s -verify-machineinstrs -o - -regbankselect-greedy | FileCheck %s
+---
+name: saddo_s32_ss
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0, $sgpr1
+    ; CHECK-LABEL: name: saddo_s32_ss
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
+    ; CHECK: [[SADDO:%[0-9]+]]:sgpr(s32), [[SADDO1:%[0-9]+]]:scc(s1) = G_SADDO [[COPY]], [[COPY1]]
+    %0:_(s32) = COPY $sgpr0
+    %1:_(s32) = COPY $sgpr1
+    %2:_(s32), %3:_(s1) = G_SADDO %0, %1
+...
+
+---
+name: saddo_s32_sv
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0, $vgpr0
+    ; CHECK-LABEL: name: saddo_s32_sv
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
+    ; CHECK: [[SADDO:%[0-9]+]]:vgpr(s32), [[SADDO1:%[0-9]+]]:sgpr(s1) = G_SADDO [[COPY2]], [[COPY1]]
+    %0:_(s32) = COPY $sgpr0
+    %1:_(s32) = COPY $vgpr0
+    %2:_(s32), %3:_(s1) = G_SADDO %0, %1
+...
+
+---
+name: saddo_s32_vs
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0, $vgpr0
+    ; CHECK-LABEL: name: saddo_s32_vs
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; CHECK: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; CHECK: [[SADDO:%[0-9]+]]:vgpr(s32), [[SADDO1:%[0-9]+]]:sgpr(s1) = G_SADDO [[COPY]], [[COPY2]]
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $sgpr0
+    %2:_(s32), %3:_(s1) = G_SADDO %0, %1
+...
+
+---
+name: saddo_s32_vv
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+    ; CHECK-LABEL: name: saddo_s32_vv
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+    ; CHECK: [[SADDO:%[0-9]+]]:vgpr(s32), [[SADDO1:%[0-9]+]]:sgpr(s1) = G_SADDO [[COPY]], [[COPY1]]
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(s32), %3:_(s1) = G_SADDO %0, %1
+...
diff --git a/test/CodeGen/AMDGPU/GlobalISel/regbankselect-ssubo.mir b/test/CodeGen/AMDGPU/GlobalISel/regbankselect-ssubo.mir
new file mode 100644
index 00000000000..11cab6b89ba
--- /dev/null
+++ b/test/CodeGen/AMDGPU/GlobalISel/regbankselect-ssubo.mir
@@ -0,0 +1,69 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -march=amdgcn -mcpu=fiji -run-pass=regbankselect %s -verify-machineinstrs -o - -regbankselect-fast | FileCheck %s
+# RUN: llc -march=amdgcn -mcpu=fiji -run-pass=regbankselect %s -verify-machineinstrs -o - -regbankselect-fast | FileCheck %s
+
+---
+name: ssubo_s32_ss
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0, $sgpr1
+    ; CHECK-LABEL: name: ssubo_s32_ss
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
+    ; CHECK: [[SSUBO:%[0-9]+]]:sgpr(s32), [[SSUBO1:%[0-9]+]]:scc(s1) = G_SSUBO [[COPY]], [[COPY1]]
+    %0:_(s32) = COPY $sgpr0
+    %1:_(s32) = COPY $sgpr1
+    %2:_(s32), %3:_(s1) = G_SSUBO %0, %1
+...
+
+---
+name: ssubo_s32_sv
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0, $vgpr0
+    ; CHECK-LABEL: name: ssubo_s32_sv
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
+    ; CHECK: [[SSUBO:%[0-9]+]]:vgpr(s32), [[SSUBO1:%[0-9]+]]:sgpr(s1) = G_SSUBO [[COPY2]], [[COPY1]]
+    %0:_(s32) = COPY $sgpr0
+    %1:_(s32) = COPY $vgpr0
+    %2:_(s32), %3:_(s1) = G_SSUBO %0, %1
+...
+
+---
+name: ssubo_s32_vs
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0, $vgpr0
+    ; CHECK-LABEL: name: ssubo_s32_vs
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; CHECK: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; CHECK: [[SSUBO:%[0-9]+]]:vgpr(s32), [[SSUBO1:%[0-9]+]]:sgpr(s1) = G_SSUBO [[COPY]], [[COPY2]]
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $sgpr0
+    %2:_(s32), %3:_(s1) = G_SSUBO %0, %1
+...
+
+---
+name: ssubo_s32_vv
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+    ; CHECK-LABEL: name: ssubo_s32_vv
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+    ; CHECK: [[SSUBO:%[0-9]+]]:vgpr(s32), [[SSUBO1:%[0-9]+]]:sgpr(s1) = G_SSUBO [[COPY]], [[COPY1]]
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(s32), %3:_(s1) = G_SSUBO %0, %1
+...
diff --git a/test/CodeGen/AMDGPU/GlobalISel/regbankselect-uaddo.mir b/test/CodeGen/AMDGPU/GlobalISel/regbankselect-uaddo.mir
new file mode 100644
index 00000000000..ae873bc61ca
--- /dev/null
+++ b/test/CodeGen/AMDGPU/GlobalISel/regbankselect-uaddo.mir
@@ -0,0 +1,69 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -march=amdgcn -mcpu=fiji -run-pass=regbankselect %s -verify-machineinstrs -o - -regbankselect-fast | FileCheck %s
+# RUN: llc -march=amdgcn -mcpu=fiji -run-pass=regbankselect %s -verify-machineinstrs -o - -regbankselect-greedy | FileCheck %s
+
+---
+name: uaddo_s32_ss
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0, $sgpr1
+    ; CHECK-LABEL: name: uaddo_s32_ss
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
+    ; CHECK: [[UADDO:%[0-9]+]]:sgpr(s32), [[UADDO1:%[0-9]+]]:scc(s1) = G_UADDO [[COPY]], [[COPY1]]
+    %0:_(s32) = COPY $sgpr0
+    %1:_(s32) = COPY $sgpr1
+    %2:_(s32), %3:_(s1) = G_UADDO %0, %1
+...
+
+---
+name: uaddo_s32_sv
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0, $vgpr0
+    ; CHECK-LABEL: name: uaddo_s32_sv
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
+    ; CHECK: [[UADDO:%[0-9]+]]:vgpr(s32), [[UADDO1:%[0-9]+]]:sgpr(s1) = G_UADDO [[COPY2]], [[COPY1]]
+    %0:_(s32) = COPY $sgpr0
+    %1:_(s32) = COPY $vgpr0
+    %2:_(s32),  %3:_(s1) = G_UADDO %0, %1
+...
+
+---
+name: uaddo_s32_vs
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0, $vgpr0
+    ; CHECK-LABEL: name: uaddo_s32_vs
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; CHECK: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; CHECK: [[UADDO:%[0-9]+]]:vgpr(s32), [[UADDO1:%[0-9]+]]:sgpr(s1) = G_UADDO [[COPY]], [[COPY2]]
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $sgpr0
+    %2:_(s32), %3:_(s1) = G_UADDO %0, %1
+...
+
+---
+name: uaddo_s32_vv
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+    ; CHECK-LABEL: name: uaddo_s32_vv
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+    ; CHECK: [[UADDO:%[0-9]+]]:vgpr(s32), [[UADDO1:%[0-9]+]]:sgpr(s1) = G_UADDO [[COPY]], [[COPY1]]
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(s32), %3:_(s1) = G_UADDO %0, %1
+...
diff --git a/test/CodeGen/AMDGPU/GlobalISel/regbankselect-usubo.mir b/test/CodeGen/AMDGPU/GlobalISel/regbankselect-usubo.mir
new file mode 100644
index 00000000000..9dfda82bf52
--- /dev/null
+++ b/test/CodeGen/AMDGPU/GlobalISel/regbankselect-usubo.mir
@@ -0,0 +1,69 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -march=amdgcn -mcpu=fiji -run-pass=regbankselect %s -verify-machineinstrs -o - -regbankselect-fast | FileCheck %s
+# RUN: llc -march=amdgcn -mcpu=fiji -run-pass=regbankselect %s -verify-machineinstrs -o - -regbankselect-greedy | FileCheck %s
+
+---
+name: usubo_s32_ss
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0, $sgpr1
+    ; CHECK-LABEL: name: usubo_s32_ss
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
+    ; CHECK: [[USUBO:%[0-9]+]]:sgpr(s32), [[USUBO1:%[0-9]+]]:scc(s1) = G_USUBO [[COPY]], [[COPY1]]
+    %0:_(s32) = COPY $sgpr0
+    %1:_(s32) = COPY $sgpr1
+    %2:_(s32), %3:_(s1) = G_USUBO %0, %1
+...
+
+---
+name: usubo_s32_sv
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0, $vgpr0
+    ; CHECK-LABEL: name: usubo_s32_sv
+    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[COPY]](s32)
+    ; CHECK: [[USUBO:%[0-9]+]]:vgpr(s32), [[USUBO1:%[0-9]+]]:sgpr(s1) = G_USUBO [[COPY2]], [[COPY1]]
+    %0:_(s32) = COPY $sgpr0
+    %1:_(s32) = COPY $vgpr0
+    %2:_(s32),  %3:_(s1) = G_USUBO %0, %1
+...
+
+---
+name: usubo_s32_vs
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $sgpr0, $vgpr0
+    ; CHECK-LABEL: name: usubo_s32_vs
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; CHECK: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; CHECK: [[USUBO:%[0-9]+]]:vgpr(s32), [[USUBO1:%[0-9]+]]:sgpr(s1) = G_USUBO [[COPY]], [[COPY2]]
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $sgpr0
+    %2:_(s32), %3:_(s1) = G_USUBO %0, %1
+...
+
+---
+name: usubo_s32_vv
+legalized: true
+
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+    ; CHECK-LABEL: name: usubo_s32_vv
+    ; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+    ; CHECK: [[USUBO:%[0-9]+]]:vgpr(s32), [[USUBO1:%[0-9]+]]:sgpr(s1) = G_USUBO [[COPY]], [[COPY1]]
+    %0:_(s32) = COPY $vgpr0
+    %1:_(s32) = COPY $vgpr1
+    %2:_(s32), %3:_(s1) = G_USUBO %0, %1
+...