From: Matt Arsenault <Matthew.Arsenault@amd.com>
Date: Thu, 19 Sep 2019 01:42:34 +0000 (+0000)
Subject: AMDGPU/GlobalISel: Fix RegBankSelect G_SMULH/G_UMULH pre-gfx9
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=a50f0c836c7f4f9d7debea9a53774a4a9e24d17d;p=llvm

AMDGPU/GlobalISel: Fix RegBankSelect G_SMULH/G_UMULH pre-gfx9

The scalar versions were only introduced in gfx9.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@372286 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp b/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
index a90ea152538..1c59963643b 100644
--- a/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
+++ b/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp
@@ -1748,7 +1748,6 @@ AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
 
     LLVM_FALLTHROUGH;
   }
-
   case AMDGPU::G_GEP:
   case AMDGPU::G_ADD:
   case AMDGPU::G_SUB:
@@ -1764,8 +1763,6 @@ AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
   case AMDGPU::G_SADDE:
   case AMDGPU::G_USUBE:
   case AMDGPU::G_SSUBE:
-  case AMDGPU::G_UMULH:
-  case AMDGPU::G_SMULH:
   case AMDGPU::G_SMIN:
   case AMDGPU::G_SMAX:
   case AMDGPU::G_UMIN:
@@ -1799,6 +1796,13 @@ AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
   case AMDGPU::G_INTRINSIC_TRUNC:
   case AMDGPU::G_INTRINSIC_ROUND:
     return getDefaultMappingVOP(MI);
+  case AMDGPU::G_UMULH:
+  case AMDGPU::G_SMULH: {
+    if (MF.getSubtarget<GCNSubtarget>().hasScalarMulHiInsts() &&
+        isSALUMapping(MI))
+      return getDefaultMappingSOP(MI);
+    return getDefaultMappingVOP(MI);
+  }
   case AMDGPU::G_IMPLICIT_DEF: {
     unsigned Size = MRI.getType(MI.getOperand(0).getReg()).getSizeInBits();
     OpdsMapping[0] = AMDGPU::getValueMapping(AMDGPU::SGPRRegBankID, Size);
diff --git a/lib/Target/AMDGPU/AMDGPUSubtarget.h b/lib/Target/AMDGPU/AMDGPUSubtarget.h
index c9d305a5bba..bf7cf86bc42 100644
--- a/lib/Target/AMDGPU/AMDGPUSubtarget.h
+++ b/lib/Target/AMDGPU/AMDGPUSubtarget.h
@@ -555,6 +555,10 @@ public:
     return GFX9Insts;
   }
 
+  bool hasScalarMulHiInsts() const {
+    return GFX9Insts;
+  }
+
   TrapHandlerAbi getTrapHandlerAbi() const {
     return isAmdHsaOS() ? TrapHandlerAbiHsa : TrapHandlerAbiNone;
   }
diff --git a/test/CodeGen/AMDGPU/GlobalISel/regbankselect-smulh.mir b/test/CodeGen/AMDGPU/GlobalISel/regbankselect-smulh.mir
index 4eb3cae9ef9..4401a3c8bd9 100644
--- a/test/CodeGen/AMDGPU/GlobalISel/regbankselect-smulh.mir
+++ b/test/CodeGen/AMDGPU/GlobalISel/regbankselect-smulh.mir
@@ -1,5 +1,9 @@
 # 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=tahiti -run-pass=regbankselect -regbankselect-fast -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX6 %s
+# RUN: llc -march=amdgcn -mcpu=tahiti -run-pass=regbankselect -regbankselect-greedy -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX6 %s
+
+# RUN: llc -march=amdgcn -mcpu=gfx900 -run-pass=regbankselect -regbankselect-fast -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX9 %s
+# RUN: llc -march=amdgcn -mcpu=gfx900 -run-pass=regbankselect -regbankselect-greedy -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX9 %s
 
 ---
 name: smulh_s32_ss
@@ -8,10 +12,16 @@ legalized: true
 body: |
   bb.0:
     liveins: $sgpr0, $sgpr1
-    ; CHECK-LABEL: name: smulh_s32_ss
-    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
-    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
-    ; CHECK: [[SMULH:%[0-9]+]]:sgpr(s32) = G_SMULH [[COPY]], [[COPY1]]
+
+    ; GFX6-LABEL: name: smulh_s32_ss
+    ; GFX6: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; GFX6: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
+    ; GFX6: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; GFX6: [[SMULH:%[0-9]+]]:vgpr(s32) = G_SMULH [[COPY]], [[COPY2]]
+    ; GFX9-LABEL: name: smulh_s32_ss
+    ; GFX9: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
+    ; GFX9: [[SMULH:%[0-9]+]]:sgpr(s32) = G_SMULH [[COPY]], [[COPY1]]
     %0:_(s32) = COPY $sgpr0
     %1:_(s32) = COPY $sgpr1
     %2:_(s32) = G_SMULH %0, %1
@@ -24,10 +34,15 @@ legalized: true
 body: |
   bb.0:
     liveins: $sgpr0, $vgpr0
-    ; CHECK-LABEL: name: smulh_s32_sv
-    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
-    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
-    ; CHECK: [[SMULH:%[0-9]+]]:vgpr(s32) = G_SMULH [[COPY]], [[COPY1]]
+
+    ; GFX6-LABEL: name: smulh_s32_sv
+    ; GFX6: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; GFX6: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; GFX6: [[SMULH:%[0-9]+]]:vgpr(s32) = G_SMULH [[COPY]], [[COPY1]]
+    ; GFX9-LABEL: name: smulh_s32_sv
+    ; GFX9: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; GFX9: [[SMULH:%[0-9]+]]:vgpr(s32) = G_SMULH [[COPY]], [[COPY1]]
     %0:_(s32) = COPY $sgpr0
     %1:_(s32) = COPY $vgpr0
     %2:_(s32) = G_SMULH %0, %1
@@ -40,11 +55,17 @@ legalized: true
 body: |
   bb.0:
     liveins: $sgpr0, $vgpr0
-    ; CHECK-LABEL: name: smulh_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: [[SMULH:%[0-9]+]]:vgpr(s32) = G_SMULH [[COPY]], [[COPY2]]
+
+    ; GFX6-LABEL: name: smulh_s32_vs
+    ; GFX6: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; GFX6: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; GFX6: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; GFX6: [[SMULH:%[0-9]+]]:vgpr(s32) = G_SMULH [[COPY]], [[COPY2]]
+    ; GFX9-LABEL: name: smulh_s32_vs
+    ; GFX9: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; GFX9: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; GFX9: [[SMULH:%[0-9]+]]:vgpr(s32) = G_SMULH [[COPY]], [[COPY2]]
     %0:_(s32) = COPY $vgpr0
     %1:_(s32) = COPY $sgpr0
     %2:_(s32) = G_SMULH %0, %1
@@ -57,10 +78,15 @@ legalized: true
 body: |
   bb.0:
     liveins: $vgpr0, $vgpr1
-    ; CHECK-LABEL: name: smulh_s32_vv
-    ; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
-    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
-    ; CHECK: [[SMULH:%[0-9]+]]:vgpr(s32) = G_SMULH [[COPY]], [[COPY1]]
+
+    ; GFX6-LABEL: name: smulh_s32_vv
+    ; GFX6: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; GFX6: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+    ; GFX6: [[SMULH:%[0-9]+]]:vgpr(s32) = G_SMULH [[COPY]], [[COPY1]]
+    ; GFX9-LABEL: name: smulh_s32_vv
+    ; GFX9: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+    ; GFX9: [[SMULH:%[0-9]+]]:vgpr(s32) = G_SMULH [[COPY]], [[COPY1]]
     %0:_(s32) = COPY $vgpr0
     %1:_(s32) = COPY $vgpr1
     %2:_(s32) = G_SMULH %0, %1
diff --git a/test/CodeGen/AMDGPU/GlobalISel/regbankselect-umulh.mir b/test/CodeGen/AMDGPU/GlobalISel/regbankselect-umulh.mir
index d801149e7a1..f62cd4cd469 100644
--- a/test/CodeGen/AMDGPU/GlobalISel/regbankselect-umulh.mir
+++ b/test/CodeGen/AMDGPU/GlobalISel/regbankselect-umulh.mir
@@ -1,5 +1,9 @@
 # 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=tahiti -run-pass=regbankselect -regbankselect-fast -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX6 %s
+# XUN: llc -march=amdgcn -mcpu=tahiti -run-pass=regbankselect -regbankselect-greedy -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX6 %s
+
+# RUN: llc -march=amdgcn -mcpu=gfx900 -run-pass=regbankselect -regbankselect-fast -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX9 %s
+# XUN: llc -march=amdgcn -mcpu=gfx900 -run-pass=regbankselect -regbankselect-greedy -verify-machineinstrs -o - %s | FileCheck -check-prefix=GFX9 %s
 
 ---
 name: umulh_s32_ss
@@ -8,10 +12,16 @@ legalized: true
 body: |
   bb.0:
     liveins: $sgpr0, $sgpr1
-    ; CHECK-LABEL: name: umulh_s32_ss
-    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
-    ; CHECK: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
-    ; CHECK: [[UMULH:%[0-9]+]]:sgpr(s32) = G_UMULH [[COPY]], [[COPY1]]
+
+    ; GFX6-LABEL: name: umulh_s32_ss
+    ; GFX6: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; GFX6: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
+    ; GFX6: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; GFX6: [[UMULH:%[0-9]+]]:vgpr(s32) = G_UMULH [[COPY]], [[COPY2]]
+    ; GFX9-LABEL: name: umulh_s32_ss
+    ; GFX9: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr1
+    ; GFX9: [[UMULH:%[0-9]+]]:sgpr(s32) = G_UMULH [[COPY]], [[COPY1]]
     %0:_(s32) = COPY $sgpr0
     %1:_(s32) = COPY $sgpr1
     %2:_(s32) = G_UMULH %0, %1
@@ -24,10 +34,15 @@ legalized: true
 body: |
   bb.0:
     liveins: $sgpr0, $vgpr0
-    ; CHECK-LABEL: name: umulh_s32_sv
-    ; CHECK: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
-    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
-    ; CHECK: [[UMULH:%[0-9]+]]:vgpr(s32) = G_UMULH [[COPY]], [[COPY1]]
+
+    ; GFX6-LABEL: name: umulh_s32_sv
+    ; GFX6: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; GFX6: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; GFX6: [[UMULH:%[0-9]+]]:vgpr(s32) = G_UMULH [[COPY]], [[COPY1]]
+    ; GFX9-LABEL: name: umulh_s32_sv
+    ; GFX9: [[COPY:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; GFX9: [[UMULH:%[0-9]+]]:vgpr(s32) = G_UMULH [[COPY]], [[COPY1]]
     %0:_(s32) = COPY $sgpr0
     %1:_(s32) = COPY $vgpr0
     %2:_(s32) = G_UMULH %0, %1
@@ -40,11 +55,17 @@ legalized: true
 body: |
   bb.0:
     liveins: $sgpr0, $vgpr0
-    ; CHECK-LABEL: name: umulh_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: [[UMULH:%[0-9]+]]:vgpr(s32) = G_UMULH [[COPY]], [[COPY2]]
+
+    ; GFX6-LABEL: name: umulh_s32_vs
+    ; GFX6: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; GFX6: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; GFX6: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; GFX6: [[UMULH:%[0-9]+]]:vgpr(s32) = G_UMULH [[COPY]], [[COPY2]]
+    ; GFX9-LABEL: name: umulh_s32_vs
+    ; GFX9: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:sgpr(s32) = COPY $sgpr0
+    ; GFX9: [[COPY2:%[0-9]+]]:vgpr(s32) = COPY [[COPY1]](s32)
+    ; GFX9: [[UMULH:%[0-9]+]]:vgpr(s32) = G_UMULH [[COPY]], [[COPY2]]
     %0:_(s32) = COPY $vgpr0
     %1:_(s32) = COPY $sgpr0
     %2:_(s32) = G_UMULH %0, %1
@@ -57,10 +78,15 @@ legalized: true
 body: |
   bb.0:
     liveins: $vgpr0, $vgpr1
-    ; CHECK-LABEL: name: umulh_s32_vv
-    ; CHECK: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
-    ; CHECK: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
-    ; CHECK: [[UMULH:%[0-9]+]]:vgpr(s32) = G_UMULH [[COPY]], [[COPY1]]
+
+    ; GFX6-LABEL: name: umulh_s32_vv
+    ; GFX6: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; GFX6: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+    ; GFX6: [[UMULH:%[0-9]+]]:vgpr(s32) = G_UMULH [[COPY]], [[COPY1]]
+    ; GFX9-LABEL: name: umulh_s32_vv
+    ; GFX9: [[COPY:%[0-9]+]]:vgpr(s32) = COPY $vgpr0
+    ; GFX9: [[COPY1:%[0-9]+]]:vgpr(s32) = COPY $vgpr1
+    ; GFX9: [[UMULH:%[0-9]+]]:vgpr(s32) = G_UMULH [[COPY]], [[COPY1]]
     %0:_(s32) = COPY $vgpr0
     %1:_(s32) = COPY $vgpr1
     %2:_(s32) = G_UMULH %0, %1